Enhanced Space Debris detection and monitoring using a hybrid Bi-LSTM- CNN and Bayesian Optimization

Enhanced Space Debris detection and monitoring using a hybrid Bi-LSTM- CNN and Bayesian Optimization | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Enhanced Space Debris detection and monitoring using a hybrid Bi-LSTM- CNN and Bayesian Optimization Ishaani Priyadarshini This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3968729/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Monitoring space debris is vital for ensuring the safety of space missions and satellite operations amid the increasing number of satellites and spacecraft in orbit. The study addresses this challenge by proposing a novel approach based on a hybrid Bi-LSTM-CNN architecture optimized using Bayesian Optimization. Through extensive analysis utilizing machine learning and deep learning techniques, the study develops a robust space debris detection system capable of classifying both the object type and Radar Cross Section (RCS) size. The proposed method outperforms existing approaches by demonstrating superior performance across multiple evaluation metrics, including accuracy, precision, recall, and F1 score. Moreover, the study considers the practical aspect of training time, ensuring efficiency in real-time applications. Empirical validation on real-world datasets confirms the effectiveness and efficiency of the hybrid model in accurately detecting and predicting space debris types. Overall, this research significantly advances space debris monitoring capabilities, mitigating risks associated with space exploration and satellite operations, and offers comprehensive insights into potential hazards and optimizing mitigation strategies. Bayesian Optimization Deep Learning Monitoring System Space Debris Bi-LSTM-CNN Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 1. Introduction Space debris, or space junk, refers to defunct human-made objects orbiting the Earth that no longer serve any useful purpose. These objects range from spent rocket stages and defunct satellites to fragments generated by collisions or explosions in space (Svotina and Cherkasova, 2023 ). As the number of satellites and spacecraft launched into orbit increases, so does the risk of collisions with this debris, posing significant threats to space missions, satellite operations, and even human life. Monitoring space debris is essential to mitigate these risks and ensure the sustainability of space activities. The primary purpose of monitoring space debris is to track the location, trajectory, and characteristics of debris objects orbiting the Earth. This information allows space agencies, satellite operators, and other stakeholders to assess collision risks and take necessary measures to avoid potential impacts. These measures may include adjusting satellite orbits, implementing collision avoidance maneuvers, or decommissioning satellites at the end of their operational life to prevent them from becoming additional sources of debris (Ledkov and Aslanov, 2023 ) (Wang et al., 2023 ). Space debris threatens space infrastructure, including operational satellites, stations, and crewed spacecraft. Collisions with debris particles can cause catastrophic damage, losing valuable assets and disrupting vital communication, navigation, and Earth observation services. Moreover, the proliferation of debris in certain orbital regions, such as low Earth orbit (LEO), increases the likelihood of cascading collisions, known as the Kessler syndrome, where one collision triggers a chain reaction of further collisions, creating a dense cloud of debris that renders the affected region unusable for space activities. Given the critical importance of monitoring space debris to ensure the safety and sustainability of space activities, it has emerged as a significant area of focus for research and development in artificial intelligence (AI) for social good. By leveraging AI techniques, such as machine learning and deep learning, researchers aim to enhance the accuracy and efficiency of space debris monitoring systems (Viavattene et al., 2022 ). These AI-driven approaches enable the analysis of large volumes of observational data collected from ground-based sensors, telescopes, and space-based assets to identify and track debris objects more effectively. Furthermore, AI-powered space debris monitoring systems enhance operational efficiency and promote global collaboration and responsibility in space exploration. By providing real-time risk assessments and predictive analytics, these systems empower satellite operators and space agencies to manage collision risks proactively, minimizing the potential for catastrophic accidents and preserving critical space assets. This proactive approach fosters international cooperation and strengthens space governance frameworks, promoting the peaceful and sustainable use of outer space for the benefit of all humankind. In essence, by integrating AI into space debris monitoring, the study not only improves space situational awareness and enhances space traffic management but also contributes to the broader goal of ensuring the long-term sustainability and peaceful coexistence of humanity in space, exemplifying the transformative potential of AI for social good (Wu et al., 2024 ). Predicting the size of RCS (Radar Cross Section) can aid in space debris identification by providing insights into the physical characteristics and potential hazards posed by the debris. A larger RCS typically indicates a larger physical size of the debris object, which could pose greater risks to operational spacecraft or satellites due to potential collisions. By accurately predicting the size of RCS, space agencies and satellite operators can better assess the risk of collision with debris, enabling them to implement appropriate mitigation measures such as maneuvering satellites to avoid collisions or planning debris removal missions. Additionally, understanding the size distribution of RCS can inform spacecraft design considerations, such as incorporating stronger shielding to mitigate the impact of potential collisions with larger debris objects. Addressing the critical challenges in space debris monitoring requires innovative approaches that leverage the strengths of artificial intelligence (AI) techniques and optimization algorithms. A hybrid model, the Bi-LSTM CNN architecture optimized using Bayesian Optimization, has been proposed in response. This model represents a significant advancement in space debris monitoring by combining the capabilities of LSTM (Long Short-Term Memory) networks with convolutional neural networks (CNNs) and Bayesian Optimization. The LSTM (Long Short-Term Memory) layers excel at capturing temporal dependencies present in sequential data. In the case of space debris monitoring, sequential data may include time-series measurements from various sensors, such as radar or infrared sensors, capturing parameters like velocity, position, and other relevant attributes of debris objects in orbit. The LSTM component of the architecture can analyze these sequential data streams to identify patterns, trends, and anomalies indicative of potential collision risks. On the other hand, the CNN (Convolutional Neural Network) layers are adept at extracting spatial features from multidimensional data. While typically used for image data, CNNs can also be applied to numerical data with spatial structure, such as grids or matrices representing spatial relationships. In space debris monitoring, this could involve data representing the spatial distribution of debris objects in different orbital regions or the layout of sensor networks for monitoring. By integrating LSTM and CNN layers into the architecture, our proposed Bi-LSTM CNN model can effectively analyze the multidimensional numeric data encountered in space debris monitoring. Furthermore, incorporating Bayesian Optimization allows automatic tuning of hyperparameters, resulting in improved model performance and efficiency. Bayesian Optimization is particularly well-suited for optimizing complex, high-dimensional search spaces, making it an ideal choice for fine-tuning the parameters of the Bi-LSTM CNN architecture (Deng et al., 2024 ) (Khan et al., 2023 ). One of the key highlights of our approach lies in integrating a diverse set of machine learning (ML) and deep learning algorithms to assess the performance of the proposed model. Therefore, the approach is twofold: a classification is conducted to identify the type of debris and simultaneously classify the size of RCS using the same dataset. The model's capabilities across different data types and modalities have been validated by leveraging machine learning algorithms and deep learning architectures. To evaluate the performance of the Bi-LSTM CNN model, a range of metrics, including accuracy, precision, recall, and F1 score has been employed. These metrics comprehensively assess the model's ability to detect and classify space debris objects accurately. Additionally, training time has been considered a critical factor, ensuring the model is efficient and scalable for real-world deployment. The rest of the paper is organized as follows. Section 2 encompasses materials and methods wherein some related works, the methodology, and the proposed architecture have been discussed. The main emphasis is given to the BiLSTM-CNN technique and the Bayesian Optimization. Section 3 highlights the overall results and observations based on the dataset and experimental analysis. A comparative analysis depicts how the proposed work contributes to the space debris monitoring research, followed by overall observations. Section 4 concludes the study. 2. Materials and Methods In this section, there is an emphasis on some of the recent related works concerning space debris monitoring. The study also highlights the limitations of the existing approaches for identifying space debris. This is followed by a methodology section focusing on the proposed architecture and its components. Consequently, the algorithm and the pseudocode have been analyzed. 2.1 Related Works A study (Jordan et al., 2023 ) proposed a PSO-based method to estimate inertia parameters for uncooperative satellites in space debris removal. Their approach focuses on estimating the inertia of a rotating target in a torque-free environment using quaternion data from attitude observations. The authors estimate the target's symmetric inertia tensor tensor by representing the PSO solution space as a multi-dimensional vector mapping to the inertia tensor. Euler's equations propagate attitude motion to generate estimated measurements, validated against experimental data with < 1% error. However, limitations include the assumption of a torque-free environment, which is potentially unrealistic for debris removal scenarios, and reliance solely on experimental measurements for validation, which overlooks measurement errors. Further validation under diverse conditions is needed to assess the method's robustness. Another study (Ryan et al., 2023) explored machine learning (ML) for spacecraft protection against micrometeoroid and orbital debris impacts. Their study highlights ML's capacity to handle greater complexity than traditional semi-empirical models. Utilizing artificial neural networks (ANNs), support vector machines (SVMs), and extreme gradient boosting (XGBoost), the researchers illustrate ML's effectiveness in this domain. Notably, XGBoost emerges as the top-performing model. However, the study's scope is limited, relying on basic ML techniques. Future research employing advanced ML methods and comprehensive datasets could yield deeper insights into spacecraft risk assessment and protection strategies. Researchers (Wu and Rosengren, 2023) introduced an unsupervised learning approach, DBSCAN, to identify clusters of orbital debris using proper element data obtained from two-line element sets. Proper elements for debris fragments in low-Earth orbit are computed using a numerical scheme similar to the Fourier-series-based method for asteroids. To enhance the classical DBSCAN's heuristic nature, neural networks trained on known families are explored. However, it is important to acknowledge the study's limitations, including potential challenges in accurately representing complex orbital debris distributions with neural networks. Another study (Xu et al., 2023 ) conducted an analysis of active debris removal (ADR) mission planning, aiming to generate optimal debris removal plans. They established a two-layer time-dependent traveling salesman problem (TSP) mathematical model to address debris removal sequence and transfer trajectory planning. Novel machine learning-based methods were proposed for ADR mission planning, including a deep neural network (DNN)-based estimation method for optimal velocity increments and a reinforcement learning (RL)-based method for optimizing debris removal sequence and rendezvous time. Simulation results demonstrate higher estimation accuracy compared to analytical methods. However, the proposed methods may face challenges in scaling to more complex mission scenarios and require extensive computational resources for training neural networks in real-time applications. Some researchers (Cimmino et al., 2023 ) introduced a machine learning-driven regression method for estimating the ballistic coefficient in Low Earth Orbit, covering a broad range of orbital parameters. They evaluated various machine learning techniques using synthetic space catalog data and conducted sensitivity analyses on training size and measurement frequency factors. Despite the neural network achieving an 84% success rate, challenges arise in extrapolating the approach to real-world scenarios due to potential biases in the synthetic dataset and uncertainties in real data. Additionally, the method's reliance on accurate and comprehensive training data may limit its effectiveness in dynamic and evolving space environments. Another research group (Salmaso et al., 2023 ) proposed a deep learning model for predicting the re-entry of uncontrolled objects in Low Earth Orbit (LEO) based on a modified Sequence-to-Sequence architecture. Trained on average altitude profiles from Two-Line Element (TLE) data of over 400 bodies, the model introduces novel input features, including a drag-like coefficient (𝐵∗), average solar index, and area-to-mass ratio. Testing on Inter-Agency Space Debris Coordination Committee (IADC) campaign objects revealed optimal performance for bodies with similar drag-like coefficients and eccentricity distributions as the training set. However, when applied to objects with significantly different characteristics, the model's performance may be limited, highlighting potential challenges in generalizing to diverse space debris populations. A study (Guo et al., 2023 ) introduced a novel method for clustering spectral polarization data from space debris using a hybrid fuzzy C-means (FCM) algorithm model incorporating hierarchical agglomerative clustering (HAC). Validation of the algorithm's efficacy employed the Kosko subset measure formula. Characteristic parameters derived from laboratory test data created a clustering matrix. Algorithm parameters were determined by randomly selecting points in the external field. While the resulting approach achieved a high classification accuracy of 96.92% for six sample types in spectral polarization images, limitations may arise in its generalizability to more diverse or complex debris datasets. Additionally, the reliance on laboratory test data may not fully capture the variability present in real-world debris environments, potentially impacting the algorithm's performance in practical applications. Another research group (Singh et al., 2024 ) introduced a Physics Informed Neural Network (PINN) approach for estimating space debris trajectory post-collision events with active satellites. The simulation involved 8565 inelastic collisions using Two Line Element (TLE) data for 1647 Starlink and 66 Low Earth orbit Multi-Use Receiver (LEMUR) satellites. Despite comprehensive simulation and proposed velocity sampling methods, the classical optimization method, the Lagrange multiplier approach, yielded unsatisfactory state estimation due to under-determination. Alternative Deep Neural Network and PINN-based methods were developed, with PINN-based approaches demonstrating superior performance in estimating position, velocity, mass, and coefficient of restitution of space debris. However, limitations may arise in extrapolating results to highly dynamic and complex collision scenarios not fully represented in the simulation dataset. Additionally, the efficacy of the proposed methods may vary depending on the availability and quality of input data from real-world collision events. Researchers (Zhang et al., 2024 ) reviewed deep learning methods for detecting space targets and components, emphasizing their importance in advancing space missions. While effective detection and recognition of space targets using electro-optical sensors are crucial for spacecraft control, limitations exist in current research. The review systematically covers the principles and characteristics of these sensors and common synthetic methods for space target datasets. Despite summarizing recent research and addressing major issues in space target detection and segmentation, the applicability of deep learning methods may be constrained by challenges such as limited training data and variable environmental conditions in space. Another study (Pensado et al., 2023 ) explored the application of a commercial global flash Light Detection and Ranging (LiDAR) sensor in active debris removal operations, emphasizing the need for precise target positioning and orientation. While relative navigation devices like cameras or LiDAR sensors are commonly used for such missions, this study simulated data acquisition and processing from a commercial LiDAR sensor. The novelty lies in using multilayer perceptron neural networks to process LiDAR depth images for estimating the target's pose. However, limitations may arise from the complexity of accurately modeling real-world debris environments and the potential variability in LiDAR sensor performance under different conditions. In the past, various techniques have been utilized to monitor and identify space debris, each with limitations. Despite this, the potential of artificial intelligence, particularly in machine learning and deep learning, remains largely untapped. By leveraging advanced machine learning and deep learning models, the study aims to address the shortcomings of previous approaches. Thus, the study proposes adopting the BiLSTM CNN model optimized with Bayesian optimization, which promises to offer robust solutions and overcome the limitations encountered in earlier methods. 2.2 Methodology The process begins with thorough data preparation to effectively classify space debris and RCS size using machine learning. This involves gathering and preprocessing the dataset, ensuring completeness and uniformity by handling missing values, outliers, and inconsistencies. Additionally, features are carefully engineered to capture relevant information, and selections are made to enhance model performance. The dataset is split into training and test sets, maintaining balanced class distributions to prevent bias. With the data ready, diverse ML algorithms are deployed alongside deep learning algorithms and ensemble techniques. After deployment, the performance of the models is validated. Each model is trained on the training data and evaluated on the test data using various performance metrics, including accuracy, precision, recall, and F1-score. Among the extensive suite of algorithms, the proposed approach integrates BiLSTM-CNN with Bayesian Optimization. This model combines the benefits of deep learning and optimization techniques, offering robust and efficient classification performance. Figure 1 depicts the overall methodology of the process. a. Bi-LSTM-CNN Architecture This study proposes a hybrid Bi-LSTM-CNN method for classifying space debris and RCS sizes. The overall architecture comprises Bi-LSTM and CNN components, which are further integrated and result in an output layer. The BiLSTM layer processes input sequences in both forward and backward directions simultaneously. This allows the model to capture temporal dependencies from past and future time steps. Each LSTM unit maintains a memory cell to store information over time, allowing it to retain long-term dependencies. LSTM units incorporate gate mechanisms (i.e., input, forget, and output gates) to regulate the flow of information through the network, facilitating better gradient flow and addressing the vanishing/exploding gradient problem. Each LSTM unit's hidden state captures the sequence's current state, passed to subsequent time steps and layers for further processing. The CNN layer applies convolutional filters to the input numeric data to extract local patterns or features. Each filter slides over the input sequence, performing element-wise multiplications and summations to produce feature maps. After convolution, an activation function introduces non-linearity to the extracted features, enhancing the model's capacity to capture complex patterns. Max Pooling can downsample the feature maps, reducing computational complexity and spatial dimensions while retaining important features. The output of the CNN layer serves as input to the BiLSTM layer, providing it with spatial features extracted from the numeric data. These features capture spatial patterns and relationships within the input sequence. The BiLSTM layer processes the spatial features temporally, capturing sequential patterns and dependencies within the numeric sequences over time. This integration allows the model to learn both spatial and temporal representations simultaneously. The output layer typically consists of one or more fully connected layers, which map the learned features to the desired output format. An activation function (softmax) is applied to the output of the fully connected layers to produce the final predictions or outputs. During training, a loss function measures the disparity between predicted and true labels/values, guiding the optimization process. Figure 2 depicts the overall architecture of the BiLSTM-CNN model. b. Bayesian Optimization Bayesian optimization begins with an initial set of hyperparameters for the model. These hyperparameters define the model's configuration, such as learning rate, batch size, dropout rate, etc. Bayesian optimization relies on a surrogate probabilistic model (often a Gaussian Process) to approximate the objective function, which in our case is the performance metric of the BiLSTM-CNN model (e.g., accuracy, F1-score). The surrogate model estimates the model's performance for different combinations of hyperparameters based on the evaluations conducted so far. The acquisition function guides the selection of the next set of hyperparameters to evaluate. It balances exploration (trying new hyperparameter configurations) and exploitation (evaluating promising configurations). Common acquisition functions include Expected Improvement (EI), Probability of Improvement (PI), and Upper Confidence Bound (UCB). The selected set of hyperparameters is used to train and evaluate the BiLSTM-CNN model on a subset of the training data (validation set). The model's performance metric (e.g., accuracy) is computed based on its performance on the validation set. The performance metric obtained from evaluating the objective function updates the surrogate model. The surrogate model is refined based on the new data point to better approximate the objective function. Building a surrogate function and acquisition function are repeated iteratively for a predefined number of iterations or until convergence criteria are met. At each iteration, the acquisition function is used to select the next set of hyperparameters to evaluate, and the process continues until the optimal set of hyperparameters is found. Figure 3 shows the steps involved in Bayesian Optimization. 2.3 Algorithmic Analysis and Pseudocode In this section, the overall algorithm of the classification process has been analyzed along with the pseudocode. The following algorithm depicts the pseudocode for deploying Bi-LSTM-CNN with Bayesian Optimization for identifying Space debris (Fig. 4 ). The provided algorithm outlines the procedure for tuning the hyperparameters of a Bi-LSTM-CNN model using Bayesian Optimization. Initially, Bayesian Optimization is initialized, followed by defining the hyperparameter search space. An objective function assesses the model's performance based on specified metrics. The algorithm iterates through a set number of iterations, selecting hyperparameters, training the model, evaluating its performance, and updating the Bayesian Optimization process. The optimal hyperparameters are obtained once the iterations are completed, and the final model is trained using these parameters. Finally, the performance of the trained model is evaluated, and the results are outputted. This approach automates the hyperparameter tuning process, allowing for efficient exploration of the hyperparameter space to find configurations that optimize the model's performance. The overall process can be expressed in the following steps. Step 1: Initialize the Bayesian optimization algorithm. Step 2: Define the range of hyperparameters to be optimized. Step 3: Define the performance metric (e.g., accuracy, F1-score) used to evaluate the model. Step 4: Bayesian Optimization Loop Use the Bayesian optimization algorithm to select the next set of hyperparameters to evaluate. Train the BiLSTM-CNN model using the selected hyperparameters. Evaluate the model's performance on a validation set using the defined objective function. Update the Bayesian optimization algorithm with the hyperparameters and corresponding model performance. Step 5: Once the optimization loop is complete, obtain the optimal set of hyperparameters. Step 6: Train the final BiLSTM-CNN model using the optimal hyperparameters obtained from Bayesian optimization. Step 7: Evaluate the performance of the final model on a separate test set. Step 8: Output the final model performance and any other relevant results. 3. Results and Observations This section discusses the study's dataset and the evaluation parameters for validating model performance. This is followed by an extensive analysis using diverse machine learning and deep learning algorithms. The overall performance is compared with some previous research works to make observations. 3.1 Dataset The dataset "Satellites and Debris in Earth's Orbit," sourced from Kaggle, was obtained through an API provided by space-track.org. Comprising 14,372 rows and 40 columns, it encompasses attributes such as object name, ID, country code, and launch date. The target variable, object type, categorizes entities into payload, debris, rocket body, and TBA (to be announced). The data will be partitioned into 80% training and 20% test datasets for our experimental analysis. A significant imbalance in the dataset was noticed during the initial exploratory analysis. Figure 5 illustrates the overall distribution of the data. Notably, 58.7% of the data is labeled as debris, while 34.4% corresponds to the payload. Rocket body data accounts for 5.2%, with the remaining unnamed data. To mitigate bias, the study applies sampling techniques to balance the data before conducting a comprehensive analysis. Similarly, the RCS size can be classified as small, large, and medium. 3.2 Evaluation Parameters The following evaluation parameters have been considered for analyzing the model performance Accuracy: Accuracy represents the proportion of correctly classified instances among the total number of instances. It gives an overall measure of how well the model performs across all classes. A higher accuracy indicates that the model is making fewer mistakes in its predictions. Precision: Precision measures the proportion of true positive predictions (correctly predicted positive instances) out of all instances predicted as positive. It indicates the accuracy of positive predictions and is particularly relevant when the cost of false positives is high. A high precision means that the model is good at avoiding false positives. Recall: Recall, also known as sensitivity or true positive rate, measures the proportion of true positive predictions out of all actual positive instances. It indicates the model's ability to identify all relevant instances and is crucial when the cost of false negatives is high. A high recall means that the model is good at capturing positive instances. F-1 score: The F-1 score is the harmonic mean of precision and recall. It provides a single score that balances both precision and recall. It's useful when an uneven class distribution or false positives and negatives have different costs. A higher F-1 score indicates better overall performance, balancing precision and recall. Time to run (in seconds): Time to run refers to the time it takes for the model to train on the given data and make predictions. Considering this parameter, especially for large datasets or complex models, is essential as it impacts the model's scalability and efficiency. Minimizing the time to run is often desirable, but balancing it with achieving satisfactory performance metrics is crucial. 3.3 Experimental Analysis A diverse set of algorithms, including classical machine learning techniques, deep learning techniques, and ensembles, has been considered for the experimental analysis. The study deploys logistic regression (LR), K-Nearest Neighbors (KNN), Gaussian Naive Bayes (GNB), Decision Trees (DT), Random Forests (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), Light Gradient Boosting (LGB), Support Vector Machines (SVM), Artificial Neural Networks (ANN), Feed Forward Neural Networks (FFNN), Long Short Term Memory (LSTM), Convolutional Neural Networks (CNN), and CNN-LSTM models on the dataset for classification. The performance of the proposed method, bi-LSTM-CNN with Bayesian Optimization, is compared against the diverse set of algorithms using evaluation metrics. The first phase of our study involves deploying classification models for identifying object types in the dataset, while the second phase involves identifying the RCS size (small, medium, large). Table 1 Performance of machine learning models for identifying Object Type Algorithm Accuracy Precision Recall F-1 Score Time (s) LR 0.8744 0.8109 0.8246 0.81 1.332 KNN 0.8515 0.7568 0.8278 0.78 0.870 GNB 0.8492 0.8002 0.8384 0.81 1.001 DT 0.9132 0.8422 0.8625 0.85 1.114 RF 0.9439 0.8567 0.8897 0.86 1.782 GB 0.9287 0.8018 0.9208 0.86 3.654 XGB 0.9436 0.8824 0.9118 0.89 1.161 LGB 0.9222 0.8505 0.9046 0.87 3.112 SVM 0.9328 0.8646 0.9108 0.88 8.002 ANN 0.9208 0.8104 0.8755 0.84 6.038 FFNN 0.9451 0.9036 0.9228 0.91 14.732 LSTM 0.9344 0.8877 0.9105 0.89 12.343 CNN 0.9678 0.9412 0.9478 0.94 20.861 CNN-LSTM 0.9604 0.9238 0.9558 0.93 28.446 Bi-LSTM- CNN 0.9864 0.9123 0.9766 0.94 24.889 Bi-LSTM- CNN with Bayesian Optimization 0.9916 0.9604 0.9824 0.97 47.662 Table 1 illustrates the performance evaluation of a comprehensive array of algorithms utilized in the analysis. The evaluation criteria include accuracy, precision, recall, F1 score, and the duration taken for model deployment. Across most algorithms, satisfactory performance is observed for multiclass classification. Logistic regression, K-nearest neighbor, and Gaussian naive Bayes exhibit relatively lower accuracy levels at 87%, 85%, and 84%, respectively. However, they boast shorter training times of 1.332s, 0.87s, and 1.001s, respectively. Conversely, neural networks like CNN and CNN-LSTM demonstrate excellent performance with accuracy rates of 96%, accompanied by longer training times of 20.86 seconds and 28.446 seconds, respectively. Bi-LSTM-CNN surpasses others in accuracy with 98% but requires a longer training duration of 24.889s. Notably, our proposed model, Bi-LSTM-CNN with Bayesian Optimization (Bi-LSTM-CNN-BO), achieves the highest accuracy of 99.16%, albeit with a training time of 47.66s. The rise in training time may be attributed to several reasons. BiLSTM-CNN models combine the complexities of both Bidirectional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNNs), requiring substantial computational resources to train. Bayesian optimization explores a wide range of hyperparameters to find the optimal configuration for the model, which involves training and evaluating the model multiple times, increasing the overall training time. Moreover, Bayesian optimization may require many iterations before converging to the optimal set of hyperparameters, prolonging the training process as it iteratively refines the model configuration. Figure 6 and Fig. 7 depict the performance evaluation with respect to accuracy and time (seconds) for all the deployed algorithms. Table 2 Performance of machine learning models for identifying RCS Size Algorithm Accuracy Precision Recall F-1 Score Time (s) LR 0.9333 0.9104 0.9040 0.90 0.414 KNN 0.9122 0.8566 0.8778 0.86 0.668 GNB 0.8989 0.8678 0.8944 0.87 1.003 DT 0.9111 0.8222 0.8899 9.85 1.182 RF 0.9389 0.7584 0.7784 0.76 2.324 GB 0.9144 0.8333 0.8541 0.84 2.882 XGB 0.9123 0.8312 0.8717 0.85 2.006 LGB 0.9189 0.8678 0.8802 0.87 2.267 SVM 0.9465 0.8686 0.8993 0.87 7.067 ANN 0.9499 0.9129 0.9118 0.91 5.114 FFNN 0.9312 0.8504 0.8894 0.86 11.998 LSTM 0.9552 0.8436 0.8902 0.86 17.614 CNN 0.9799 0.8211 0.8894 0.85 23.889 CNN-LSTM 0.9804 0.9004 0.9122 0.90 38.775 Bi-LSTM- CNN 0.9899 0.9234 0.9403 0.93 45.222 Bi-LSTM- CNN with Bayesian Optimization 0.9998 0.9666 0.9438 0.95 53.023 Table 2 illustrates the performance evaluation of a comprehensive array of algorithms utilized in the analysis. The evaluation criteria include accuracy, precision, recall, F1 score, and the duration taken for model deployment. Across most algorithms, satisfactory performance is observed for multiclass classification. Logistic regression, K-nearest neighbor, and Gaussian naive Bayes exhibit relatively lower accuracy levels at 93%, 91%, and 89%, respectively. However, they boast shorter training times of 0.414s, 0.668s, and 1.003s, respectively. Conversely, neural networks like CNN and CNN-LSTM demonstrate excellent performance with accuracy rates of 97.99% and 98.04%, accompanied by longer training times of 23.889s and 38.775s, respectively. Bi-LSTM-CNN surpasses others in accuracy with 98.99% but requires a longer training duration of 45.222s. Notably, our proposed model, Bi-LSTM-CNN with Bayesian Optimization (Bi-LSTM-CNN-BO), achieves the highest accuracy of 99.98%, albeit with a training time of 53.023 seconds. Figure 8 and Fig. 9 depict the performance evaluation with respect to accuracy and time (seconds) for all the deployed algorithms. 3.4 Comparative Analysis In this segment, a comparative examination between our proposed approach and several existing methods for identifying space debris has been conducted. Table 3 outlines the results of this comparative analysis. Table 3 Comparative Analysis of proposed work with relevant works Author and Year Proposed Work Methodology/ Parameters Results Jordan et al, 2023 Space Debris Removal PSO-based method to estimate inertia parameters for uncooperative satellites The proposed method exhibits less than 1% error Ryan et al, 2023] Orbital debris impact risk assessments ML algorithms like ANN, SVM, XGBoost, etc XGBoost shows the best performance with an accuracy 97.4% Guo et al, 2023 Clustering spectral polarization data from space debris Hybrid fuzzy C-means (FCM) algorithm model incorporating hierarchical agglomerative clustering (HAC) Accuracy of 96.92% Qashoa and Lee, 2023 Classifying low-orbit space objects Algorithms like SVM, LSTM LSTM exhibits an accuracy of 92% Zhao et al., 2023 Task Allocation for Space Debris Removal Improved Particle Swarm Optimization (PSO) Algorithm Improved PSO is 22.8% faster than traditional PSO Our Proposed Work Monitoring and Classifying Space Debris Extensive ML algorithms, proposed Bi-LSTM-CNN, and Bayesian Optimization The proposed method archives accuracy of 99.16% and 99.98% 3.5 Observations Several key points have been observed based on the experimental and comparative analysis. Previous research in space exploration leveraging AI has been extensive but focused primarily on specific methodologies. This study conducts a thorough analysis employing diverse algorithms, contributing to the field of space debris analysis. Machine learning algorithms (classical, ensembles, neural networks) are assessed based on key performance metrics such as precision, accuracy, recall, F1 score, and deployment time. A robust Bi-LSTM-CNN model optimized with Bayesian techniques emerges as the top performer across multiple analyses, showcasing its efficacy in space debris identification and classification. The initial analysis involves classifying debris based on object type, including debris, payload, and rocket bodies. In contrast, the subsequent analysis focuses on identifying RCS size, a critical factor for understanding debris characteristics and potential hazards. Comparative analysis reveals that this approach achieves satisfactory accuracy compared to previous works, highlighting its effectiveness in identifying and monitoring space debris. 3.6 Limitations Some limitations of the study are as follows: The study utilized a specific dataset for analysis, which may not fully represent the entire spectrum of space debris characteristics and behaviors. This could affect the generalizability of the findings. The study assumes the quality and reliability of the dataset used for analysis. However, data inconsistencies, missing values, or errors could potentially introduce biases and affect the robustness of the results. The complexity of the Bi-LSTM-CNN model with Bayesian optimization may lead to challenges in interpretability and scalability. Additionally, the computational resources required to train and deploy this model may limit its practical applicability in certain contexts. External factors such as changes in space policies, advancements in technology, or unforeseen events could influence the dynamics of space debris and render the study findings outdated or less relevant over time. 4. Conclusion The comprehensive analysis conducted in this study sheds light on the intricate task of space debris identification and classification, presenting valuable insights and potential avenues for further research. Through the utilization of diverse machine learning algorithms, including logistic regression, K-nearest neighbors, Gaussian naive Bayes, and neural networks like CNN and CNN-LSTM, the performance of various models in classifying space debris types and predicting Radar Cross Section (RCS) sizes has been examined. Additionally, introducing a robust Bi-LSTM-CNN model optimized with Bayesian techniques showcased remarkable accuracy in both classification tasks, albeit with a longer training time attributed to the model's complexity. The findings underscore the importance of accurately identifying and monitoring space debris, which poses significant risks to space missions and satellite operations. Despite the satisfactory performance achieved by the proposed model, it is crucial to acknowledge the study's limitations. The reliance on a specific dataset may limit the generalizability of the results, and potential data quality issues could introduce biases. Future research endeavors should address these limitations by incorporating larger and more diverse datasets, ensuring data quality and reliability, and exploring advanced techniques to enhance model performance and efficiency. Additionally, efforts should be directed toward integrating real-time monitoring systems and predictive analytics to mitigate the growing threat of space debris and safeguard space infrastructure. In conclusion, while the study marks a significant step towards advancing space debris analysis, it also highlights the need for continuous innovation and collaboration within the scientific community to address the challenges posed by space debris and ensure the sustainability of space exploration endeavors. Declarations Author Contribution I.P. wrote the article, performed analysis, proofreadsingle author contribution Funding Declaration This research project received no external funding. The author declares that they have not received any financial support or funding from any organization, institution, or individual for this research. Data Availability Declaration The datasets cited in this manuscript are publicly available and can be accessed from the original sources as referenced in the text. No additional data were generated or analyzed during the current study Competing Interest declaration The authors declare that they have no competing interests regarding the publication of this manuscript References Svotina, V. V., & Cherkasova, М. V. (2023). Space debris removal–Review of technologies and techniques. Flexible or virtual connection between space debris and service spacecraft. Acta Astronautica, 204, 840-853. Ledkov, A. S., & Aslanov, V. S. (2023). Active space debris removal by ion multi-beam shepherd spacecraft. Acta Astronautica, 205, 247-257. Wang, C., Chen, D., Liao, W., & Liang, Z. (2023). Autonomous obstacle avoidance strategies in the mission of large space debris removal using potential function. Advances in Space Research, 72(7), 2860-2873. Viavattene, G., Devereux, E., Snelling, D., Payne, N., Wokes, S., & Ceriotti, M. (2022). Design of multiple space debris removal missions using machine learning. Acta Astronautica, 193, 277-286. Wu, Y., Peng, K. L., Yao, Y., & Guo, Y. (2024). Sustainable Space Travel: What Can We Do in Education from Economic and Environmental Perspectives?. Sustainability, 16(2), 684. Deng, S., Su, J., Zhu, Y., Yu, Y., & Xiao, C. (2024). Forecasting carbon price trends based on an interpretable light gradient boosting machine and Bayesian optimization. Expert Systems with Applications, 242, 122502. Khan, M. A., Arshad, H., Khan, W. Z., Alhaisoni, M., Tariq, U., Hussein, H. S., ... & Elashry, A. (2023). HGRBOL2: human gait recognition for biometric application using Bayesian optimization and extreme learning machine. Future Generation Computer Systems, 143, 337-348. Jordan, J., Posada, D., Zuehlke, D., Nocerino, A., Fontdegloria, P., John, S., ... & Henderson, T. (2023, March). Estimation of Uncooperative Satellite Inertia Parameters for Space Debris Removal Using Particle Swarm Optimization. In 2023 IEEE Aerospace Conference (pp. 1-13). IEEE. Ryan, S., Sushma, N. M., Le, H., Kumar, A. A., Berk, J., Nguyen, T. M., ... & Venkatesh, S. (2023). The application of machine learning in micrometeoroid and orbital debris impact protection and risk assessment for spacecraft. International Journal of Impact Engineering, 181, 104727. Wu, D., & Rosengren, A. J. (2023). An investigation on space debris of unknown origin using proper elements and neural networks. Celestial Mechanics and Dynamical Astronomy, 135(4), 44. Xu, Y., Liu, X., He, R., Zhu, Y., Zuo, Y., & He, L. (2023). Active Debris Removal Mission Planning Method Based on Machine Learning. Mathematics, 11(6), 1419. Cimmino, N., Opromolla, R., & Fasano, G. (2023). Machine learning-based approach for ballistic coefficient estimation of resident space objects in LEO. Advances in Space Research, 71(12), 5007-5025. Salmaso, F., Trisolini, M., & Colombo, C. (2023). A machine learning and feature engineering approach for the prediction of the uncontrolled re-entry of space objects. Aerospace, 10(3), 297. Guo, F., Zhu, J., Huang, L., Li, H., Deng, J., Jiang, H., & Hou, X. (2023). Enhancing Spatial Debris Material Classifying through a Hierarchical Clustering-Fuzzy C-Means Integration Approach. Applied Sciences, 13(8), 4754. Singh, G., Kumar, V., Buduru, A. B., & Biswas, S. K. (2024). Tracking an untracked space debris after an inelastic collision using physics informed neural network. Scientific Reports, 14(1), 3350. Zhang, H., Zhang, Y., Feng, Q., & Zhang, K. (2024). Review of Machine-Learning Approaches for Object and Component Detection in Space Electro-optical Satellites. International Journal of Aeronautical and Space Sciences, 25(1), 277-292. Pensado, E. A., de Santos, L. M. G., Sanjurjo-Rivo, M., & Jorge, H. G. (2023). Deep Learning Based Target Pose Estimation Using LiDAR Measurements in Active Debris Removal Operations. IEEE Transactions on Aerospace and Electronic Systems. Qashoa, R., & Lee, R. (2023). Classification of Low Earth Orbit (LEO) Resident Space Objects’(RSO) Light Curves Using a Support Vector Machine (SVM) and Long Short-Term Memory (LSTM). Sensors, 23(14), 6539. Zhao, Y., Wu, P., Yu, H., & Tong, Y. (2023, October). Task Allocation for Space Debris Removal based on Improved Particle Swarm Optimization Algorithm. In 2023 International Conference on Cyber-Physical Social Intelligence (ICCSI) (pp. 399-404). IEEE. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3968729","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":275898889,"identity":"83209a4d-c883-4b7d-9237-07a2a73f2f05","order_by":0,"name":"Ishaani Priyadarshini","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7ElEQVRIiWNgGAWjYBACNgYeECUhZ8DAwMzAUAAWNCBKizFEiwERWhggWhgSNxCthY//7MEPP3dYpG9n731s8MHAJrGBvXmbBF6HSeQlS/aekcjd2XPcOHGGQVpiA8+xMgJaeAwkeNskcjfcSGM+zGNwOLFBIscMvxb+M8Y//7ZJpBvcf8Z8+I/B/8QG+TcEtDDkmEkDbUkwuMHGnMxgcABoCw8BLUBnWMu2SRju7EljNuwxSDZu40krtsCnRb7/jPHNt2118ubsx5glflTYyfazH954A58WDODYRpJyELAnWccoGAWjYBQMewAAF09BJpBk72QAAAAASUVORK5CYII=","orcid":"","institution":"University of California, Berkeley","correspondingAuthor":true,"prefix":"","firstName":"Ishaani","middleName":"","lastName":"Priyadarshini","suffix":""}],"badges":[],"createdAt":"2024-02-19 03:22:41","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3968729/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3968729/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":52043281,"identity":"e121b2cf-4611-41b6-a101-368a5de18369","added_by":"auto","created_at":"2024-03-05 18:55:41","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":185829,"visible":true,"origin":"","legend":"\u003cp\u003eSpace Debris Classification Methodology\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/74b9b6b8fb7e0de9c3787052.png"},{"id":52043277,"identity":"2de404c4-6d20-490d-a5e8-951d9c9d7bec","added_by":"auto","created_at":"2024-03-05 18:55:40","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":183273,"visible":true,"origin":"","legend":"\u003cp\u003eProposed Bi-LSTM-CNN Architecture\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/fb8d36a9d7a9096a56b00a13.png"},{"id":52043278,"identity":"1e1a80a5-6320-4476-b504-c82caf45bd42","added_by":"auto","created_at":"2024-03-05 18:55:40","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":234579,"visible":true,"origin":"","legend":"\u003cp\u003eBayesian Optimization steps\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/e09048048155991d2cda5c6a.png"},{"id":52043283,"identity":"5be75b5d-0d6c-4a0e-9ed0-696d2fd09b1d","added_by":"auto","created_at":"2024-03-05 18:55:41","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":704703,"visible":true,"origin":"","legend":"\u003cp\u003eAlgorithm for Bi-LSTM-CNN with Bayesian Optimization\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/8d939c0ed31c2c9906768113.png"},{"id":52043276,"identity":"ea135a60-d276-493e-b09d-cb0e7997b07f","added_by":"auto","created_at":"2024-03-05 18:55:40","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":51284,"visible":true,"origin":"","legend":"\u003cp\u003eData Distribution for Object Type\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/948f7768089070147b7b2dae.png"},{"id":52043280,"identity":"9ef05431-c509-41f2-857a-82971cba7172","added_by":"auto","created_at":"2024-03-05 18:55:41","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":47985,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Evaluation for Object Type (Accuracy)\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/5cd422eea2164f65b27beca7.png"},{"id":52043279,"identity":"d4803344-5d54-4003-a267-9bbf3b3922d2","added_by":"auto","created_at":"2024-03-05 18:55:41","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":75164,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Evaluation for Object Type (Time)\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/1728686c45b37aed37c4df94.png"},{"id":52043282,"identity":"c4172324-af12-4cbc-b907-b56cfa72eaa1","added_by":"auto","created_at":"2024-03-05 18:55:41","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":47813,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Evaluation for RCS Size (Accuracy)\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/40b61ce2ae034ee4c347f890.png"},{"id":52043284,"identity":"9ba100c4-c4c6-4dad-9bbd-c5f81b539382","added_by":"auto","created_at":"2024-03-05 18:55:41","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":75993,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Evaluation for RCS Size (Time)\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/eaa7da7fdc210788fa5872ab.png"},{"id":52089951,"identity":"a192c0ff-2a69-4fbd-8004-40101e972c37","added_by":"auto","created_at":"2024-03-06 13:46:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1217730,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3968729/v1/eadb069c-1c16-4692-8d31-27b97f5db5eb.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Enhanced Space Debris detection and monitoring using a hybrid Bi-LSTM- CNN and Bayesian Optimization","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eSpace debris, or space junk, refers to defunct human-made objects orbiting the Earth that no longer serve any useful purpose. These objects range from spent rocket stages and defunct satellites to fragments generated by collisions or explosions in space (Svotina and Cherkasova, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). As the number of satellites and spacecraft launched into orbit increases, so does the risk of collisions with this debris, posing significant threats to space missions, satellite operations, and even human life. Monitoring space debris is essential to mitigate these risks and ensure the sustainability of space activities. The primary purpose of monitoring space debris is to track the location, trajectory, and characteristics of debris objects orbiting the Earth. This information allows space agencies, satellite operators, and other stakeholders to assess collision risks and take necessary measures to avoid potential impacts. These measures may include adjusting satellite orbits, implementing collision avoidance maneuvers, or decommissioning satellites at the end of their operational life to prevent them from becoming additional sources of debris (Ledkov and Aslanov, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) (Wang et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Space debris threatens space infrastructure, including operational satellites, stations, and crewed spacecraft. Collisions with debris particles can cause catastrophic damage, losing valuable assets and disrupting vital communication, navigation, and Earth observation services. Moreover, the proliferation of debris in certain orbital regions, such as low Earth orbit (LEO), increases the likelihood of cascading collisions, known as the Kessler syndrome, where one collision triggers a chain reaction of further collisions, creating a dense cloud of debris that renders the affected region unusable for space activities. Given the critical importance of monitoring space debris to ensure the safety and sustainability of space activities, it has emerged as a significant area of focus for research and development in artificial intelligence (AI) for social good. By leveraging AI techniques, such as machine learning and deep learning, researchers aim to enhance the accuracy and efficiency of space debris monitoring systems (Viavattene et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). These AI-driven approaches enable the analysis of large volumes of observational data collected from ground-based sensors, telescopes, and space-based assets to identify and track debris objects more effectively. Furthermore, AI-powered space debris monitoring systems enhance operational efficiency and promote global collaboration and responsibility in space exploration. By providing real-time risk assessments and predictive analytics, these systems empower satellite operators and space agencies to manage collision risks proactively, minimizing the potential for catastrophic accidents and preserving critical space assets. This proactive approach fosters international cooperation and strengthens space governance frameworks, promoting the peaceful and sustainable use of outer space for the benefit of all humankind. In essence, by integrating AI into space debris monitoring, the study not only improves space situational awareness and enhances space traffic management but also contributes to the broader goal of ensuring the long-term sustainability and peaceful coexistence of humanity in space, exemplifying the transformative potential of AI for social good (Wu et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Predicting the size of RCS (Radar Cross Section) can aid in space debris identification by providing insights into the physical characteristics and potential hazards posed by the debris. A larger RCS typically indicates a larger physical size of the debris object, which could pose greater risks to operational spacecraft or satellites due to potential collisions. By accurately predicting the size of RCS, space agencies and satellite operators can better assess the risk of collision with debris, enabling them to implement appropriate mitigation measures such as maneuvering satellites to avoid collisions or planning debris removal missions. Additionally, understanding the size distribution of RCS can inform spacecraft design considerations, such as incorporating stronger shielding to mitigate the impact of potential collisions with larger debris objects.\u003c/p\u003e \u003cp\u003eAddressing the critical challenges in space debris monitoring requires innovative approaches that leverage the strengths of artificial intelligence (AI) techniques and optimization algorithms. A hybrid model, the Bi-LSTM CNN architecture optimized using Bayesian Optimization, has been proposed in response. This model represents a significant advancement in space debris monitoring by combining the capabilities of LSTM (Long Short-Term Memory) networks with convolutional neural networks (CNNs) and Bayesian Optimization. The LSTM (Long Short-Term Memory) layers excel at capturing temporal dependencies present in sequential data. In the case of space debris monitoring, sequential data may include time-series measurements from various sensors, such as radar or infrared sensors, capturing parameters like velocity, position, and other relevant attributes of debris objects in orbit. The LSTM component of the architecture can analyze these sequential data streams to identify patterns, trends, and anomalies indicative of potential collision risks. On the other hand, the CNN (Convolutional Neural Network) layers are adept at extracting spatial features from multidimensional data. While typically used for image data, CNNs can also be applied to numerical data with spatial structure, such as grids or matrices representing spatial relationships. In space debris monitoring, this could involve data representing the spatial distribution of debris objects in different orbital regions or the layout of sensor networks for monitoring. By integrating LSTM and CNN layers into the architecture, our proposed Bi-LSTM CNN model can effectively analyze the multidimensional numeric data encountered in space debris monitoring. Furthermore, incorporating Bayesian Optimization allows automatic tuning of hyperparameters, resulting in improved model performance and efficiency. Bayesian Optimization is particularly well-suited for optimizing complex, high-dimensional search spaces, making it an ideal choice for fine-tuning the parameters of the Bi-LSTM CNN architecture (Deng et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) (Khan et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). One of the key highlights of our approach lies in integrating a diverse set of machine learning (ML) and deep learning algorithms to assess the performance of the proposed model. Therefore, the approach is twofold: a classification is conducted to identify the type of debris and simultaneously classify the size of RCS using the same dataset. The model's capabilities across different data types and modalities have been validated by leveraging machine learning algorithms and deep learning architectures. To evaluate the performance of the Bi-LSTM CNN model, a range of metrics, including accuracy, precision, recall, and F1 score has been employed. These metrics comprehensively assess the model's ability to detect and classify space debris objects accurately. Additionally, training time has been considered a critical factor, ensuring the model is efficient and scalable for real-world deployment.\u003c/p\u003e \u003cp\u003eThe rest of the paper is organized as follows. Section 2 encompasses materials and methods wherein some related works, the methodology, and the proposed architecture have been discussed. The main emphasis is given to the BiLSTM-CNN technique and the Bayesian Optimization. Section 3 highlights the overall results and observations based on the dataset and experimental analysis. A comparative analysis depicts how the proposed work contributes to the space debris monitoring research, followed by overall observations. Section 4 concludes the study.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003eIn this section, there is an emphasis on some of the recent related works concerning space debris monitoring. The study also highlights the limitations of the existing approaches for identifying space debris. This is followed by a \u003cspan\u003emethodology\u003c/span\u003e section focusing on the proposed architecture and its components. Consequently, the algorithm and the pseudocode have been analyzed.\u003c/p\u003e\n\u003cdiv id=\"Sec3\"\u003e\n \u003ch2\u003e2.1 Related Works\u003c/h2\u003e\n \u003cp\u003eA study (Jordan et al., \u003cspan\u003e2023\u003c/span\u003e) proposed a PSO-based method to estimate inertia parameters for uncooperative satellites in space debris removal. Their approach focuses on estimating the inertia of a rotating target in a torque-free environment using quaternion data from attitude observations. The authors estimate the target\u0026apos;s symmetric inertia tensor tensor by representing the PSO solution space as a multi-dimensional vector mapping to the inertia tensor. Euler\u0026apos;s equations propagate attitude motion to generate estimated measurements, validated against experimental data with \u0026lt;\u0026thinsp;1% error. However, limitations include the assumption of a torque-free environment, which is potentially unrealistic for debris removal scenarios, and reliance solely on experimental measurements for validation, which overlooks measurement errors. Further validation under diverse conditions is needed to assess the method\u0026apos;s robustness. Another study (Ryan et al., 2023) explored machine learning (ML) for spacecraft protection against micrometeoroid and orbital debris impacts. Their study highlights ML\u0026apos;s capacity to handle greater complexity than traditional semi-empirical models. Utilizing artificial neural networks (ANNs), support vector machines (SVMs), and extreme gradient boosting (XGBoost), the researchers illustrate ML\u0026apos;s effectiveness in this domain. Notably, XGBoost emerges as the top-performing model. However, the study\u0026apos;s scope is limited, relying on basic ML techniques. Future research employing advanced ML methods and comprehensive datasets could yield deeper insights into spacecraft risk assessment and protection strategies. Researchers (Wu and Rosengren, 2023) introduced an unsupervised learning approach, DBSCAN, to identify clusters of orbital debris using proper element data obtained from two-line element sets. Proper elements for debris fragments in low-Earth orbit are computed using a numerical scheme similar to the Fourier-series-based method for asteroids. To enhance the classical DBSCAN\u0026apos;s heuristic nature, neural networks trained on known families are explored. However, it is important to acknowledge the study\u0026apos;s limitations, including potential challenges in accurately representing complex orbital debris distributions with neural networks. Another study (Xu et al., \u003cspan\u003e2023\u003c/span\u003e) conducted an analysis of active debris removal (ADR) mission planning, aiming to generate optimal debris removal plans. They established a two-layer time-dependent traveling salesman problem (TSP) mathematical model to address debris removal sequence and transfer trajectory planning. Novel machine learning-based methods were proposed for ADR mission planning, including a deep neural network (DNN)-based estimation method for optimal velocity increments and a reinforcement learning (RL)-based method for optimizing debris removal sequence and rendezvous time. Simulation results demonstrate higher estimation accuracy compared to analytical methods. However, the proposed methods may face challenges in scaling to more complex mission scenarios and require extensive computational resources for training neural networks in real-time applications. Some researchers (Cimmino et al., \u003cspan\u003e2023\u003c/span\u003e) introduced a machine learning-driven regression method for estimating the ballistic coefficient in Low Earth Orbit, covering a broad range of orbital parameters. They evaluated various machine learning techniques using synthetic space catalog data and conducted sensitivity analyses on training size and measurement frequency factors. Despite the neural network achieving an 84% success rate, challenges arise in extrapolating the approach to real-world scenarios due to potential biases in the synthetic dataset and uncertainties in real data. Additionally, the method\u0026apos;s reliance on accurate and comprehensive training data may limit its effectiveness in dynamic and evolving space environments. Another research group (Salmaso et al., \u003cspan\u003e2023\u003c/span\u003e) proposed a deep learning model for predicting the re-entry of uncontrolled objects in Low Earth Orbit (LEO) based on a modified Sequence-to-Sequence architecture. Trained on average altitude profiles from Two-Line Element (TLE) data of over 400 bodies, the model introduces novel input features, including a drag-like coefficient (𝐵\u0026lowast;), average solar index, and area-to-mass ratio. Testing on Inter-Agency Space Debris Coordination Committee (IADC) campaign objects revealed optimal performance for bodies with similar drag-like coefficients and eccentricity distributions as the training set. However, when applied to objects with significantly different characteristics, the model\u0026apos;s performance may be limited, highlighting potential challenges in generalizing to diverse space debris populations. A study (Guo et al., \u003cspan\u003e2023\u003c/span\u003e) introduced a novel method for clustering spectral polarization data from space debris using a hybrid fuzzy C-means (FCM) algorithm model incorporating hierarchical agglomerative clustering (HAC). Validation of the algorithm\u0026apos;s efficacy employed the Kosko subset measure formula. Characteristic parameters derived from laboratory test data created a clustering matrix. Algorithm parameters were determined by randomly selecting points in the external field. While the resulting approach achieved a high classification accuracy of 96.92% for six sample types in spectral polarization images, limitations may arise in its generalizability to more diverse or complex debris datasets. Additionally, the reliance on laboratory test data may not fully capture the variability present in real-world debris environments, potentially impacting the algorithm\u0026apos;s performance in practical applications. Another research group (Singh et al., \u003cspan\u003e2024\u003c/span\u003e) introduced a Physics Informed Neural Network (PINN) approach for estimating space debris trajectory post-collision events with active satellites. The simulation involved 8565 inelastic collisions using Two Line Element (TLE) data for 1647 Starlink and 66 Low Earth orbit Multi-Use Receiver (LEMUR) satellites. Despite comprehensive simulation and proposed velocity sampling methods, the classical optimization method, the Lagrange multiplier approach, yielded unsatisfactory state estimation due to under-determination. Alternative Deep Neural Network and PINN-based methods were developed, with PINN-based approaches demonstrating superior performance in estimating position, velocity, mass, and coefficient of restitution of space debris. However, limitations may arise in extrapolating results to highly dynamic and complex collision scenarios not fully represented in the simulation dataset. Additionally, the efficacy of the proposed methods may vary depending on the availability and quality of input data from real-world collision events. Researchers (Zhang et al., \u003cspan\u003e2024\u003c/span\u003e) reviewed deep learning methods for detecting space targets and components, emphasizing their importance in advancing space missions. While effective detection and recognition of space targets using electro-optical sensors are crucial for spacecraft control, limitations exist in current research. The review systematically covers the principles and characteristics of these sensors and common synthetic methods for space target datasets. Despite summarizing recent research and addressing major issues in space target detection and segmentation, the applicability of deep learning methods may be constrained by challenges such as limited training data and variable environmental conditions in space. Another study (Pensado et al., \u003cspan\u003e2023\u003c/span\u003e) explored the application of a commercial global flash Light Detection and Ranging (LiDAR) sensor in active debris removal operations, emphasizing the need for precise target positioning and orientation. While relative navigation devices like cameras or LiDAR sensors are commonly used for such missions, this study simulated data acquisition and processing from a commercial LiDAR sensor. The novelty lies in using multilayer perceptron neural networks to process LiDAR depth images for estimating the target\u0026apos;s pose. However, limitations may arise from the complexity of accurately modeling real-world debris environments and the potential variability in LiDAR sensor performance under different conditions.\u003c/p\u003e\n \u003cp\u003eIn the past, various techniques have been utilized to monitor and identify space debris, each with limitations. Despite this, the potential of artificial intelligence, particularly in machine learning and deep learning, remains largely untapped. By leveraging advanced machine learning and deep learning models, the study aims to address the shortcomings of previous approaches. Thus, the study proposes adopting the BiLSTM CNN model optimized with Bayesian optimization, which promises to offer robust solutions and overcome the limitations encountered in earlier methods.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\"\u003e\n \u003ch2\u003e2.2 Methodology\u003c/h2\u003e\n \u003cp\u003eThe process begins with thorough data preparation to effectively classify space debris and RCS size using machine learning. This involves gathering and preprocessing the dataset, ensuring completeness and uniformity by handling missing values, outliers, and inconsistencies. Additionally, features are carefully engineered to capture relevant information, and selections are made to enhance model performance. The dataset is split into training and test sets, maintaining balanced class distributions to prevent bias. With the data ready, diverse ML algorithms are deployed alongside deep learning algorithms and ensemble techniques. After deployment, the performance of the models is validated. Each model is trained on the training data and evaluated on the test data using various performance metrics, including accuracy, precision, recall, and F1-score. Among the extensive suite of algorithms, the proposed approach integrates BiLSTM-CNN with Bayesian Optimization. This model combines the benefits of deep learning and optimization techniques, offering robust and efficient classification performance. Figure\u0026nbsp;\u003cspan\u003e1\u003c/span\u003e depicts the overall methodology of the process.\u003c/p\u003e\n \u003cp\u003e\u003cspan\u003e\u003cem\u003ea. Bi-LSTM-CNN Architecture\u003c/em\u003e\u003cbr\u003e\u003c/span\u003e\u003c/p\u003e\n \u003cdiv\u003e\n \u003cp\u003eThis study proposes a hybrid Bi-LSTM-CNN method for classifying space debris and RCS sizes. The overall architecture comprises Bi-LSTM and CNN components, which are further integrated and result in an output layer. The BiLSTM layer processes input sequences in both forward and backward directions simultaneously. This allows the model to capture temporal dependencies from past and future time steps. Each LSTM unit maintains a memory cell to store information over time, allowing it to retain long-term dependencies. LSTM units incorporate gate mechanisms (i.e., input, forget, and output gates) to regulate the flow of information through the network, facilitating better gradient flow and addressing the vanishing/exploding gradient problem. Each LSTM unit\u0026apos;s hidden state captures the sequence\u0026apos;s current state, passed to subsequent time steps and layers for further processing. The CNN layer applies convolutional filters to the input numeric data to extract local patterns or features. Each filter slides over the input sequence, performing element-wise multiplications and summations to produce feature maps. After convolution, an activation function introduces non-linearity to the extracted features, enhancing the model\u0026apos;s capacity to capture complex patterns. Max Pooling can downsample the feature maps, reducing computational complexity and spatial dimensions while retaining important features. The output of the CNN layer serves as input to the BiLSTM layer, providing it with spatial features extracted from the numeric data. These features capture spatial patterns and relationships within the input sequence. The BiLSTM layer processes the spatial features temporally, capturing sequential patterns and dependencies within the numeric sequences over time. This integration allows the model to learn both spatial and temporal representations simultaneously. The output layer typically consists of one or more fully connected layers, which map the learned features to the desired output format. An activation function (softmax) is applied to the output of the fully connected layers to produce the final predictions or outputs. During training, a loss function measures the disparity between predicted and true labels/values, guiding the optimization process. Figure\u0026nbsp;\u003cspan\u003e2\u003c/span\u003e depicts the overall architecture of the BiLSTM-CNN model.\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003cspan\u003e\u003cem\u003eb. Bayesian Optimization\u003c/em\u003e\u003cbr\u003e\u003c/span\u003e\u003c/p\u003e\n \u003cdiv\u003e\n \u003cp\u003eBayesian optimization begins with an initial set of hyperparameters for the model. These hyperparameters define the model\u0026apos;s configuration, such as learning rate, batch size, dropout rate, etc. Bayesian optimization relies on a surrogate probabilistic model (often a Gaussian Process) to approximate the objective function, which in our case is the performance metric of the BiLSTM-CNN model (e.g., accuracy, F1-score). The surrogate model estimates the model\u0026apos;s performance for different combinations of hyperparameters based on the evaluations conducted so far. The acquisition function guides the selection of the next set of hyperparameters to evaluate. It balances exploration (trying new hyperparameter configurations) and exploitation (evaluating promising configurations). Common acquisition functions include Expected Improvement (EI), Probability of Improvement (PI), and Upper Confidence Bound (UCB). The selected set of hyperparameters is used to train and evaluate the BiLSTM-CNN model on a subset of the training data (validation set). The model\u0026apos;s performance metric (e.g., accuracy) is computed based on its performance on the validation set. The performance metric obtained from evaluating the objective function updates the surrogate model. The surrogate model is refined based on the new data point to better approximate the objective function. Building a surrogate function and acquisition function are repeated iteratively for a predefined number of iterations or until convergence criteria are met. At each iteration, the acquisition function is used to select the next set of hyperparameters to evaluate, and the process continues until the optimal set of hyperparameters is found. Figure\u0026nbsp;\u003cspan\u003e3\u003c/span\u003e shows the steps involved in Bayesian Optimization.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\"\u003e\n \u003ch2\u003e2.3 Algorithmic Analysis and Pseudocode\u003c/h2\u003e\n \u003cp\u003eIn this section, the overall algorithm of the classification process has been analyzed along with the pseudocode. The following algorithm depicts the pseudocode for deploying Bi-LSTM-CNN with Bayesian Optimization for identifying Space debris (Fig.\u0026nbsp;\u003cspan\u003e4\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe provided algorithm outlines the procedure for tuning the hyperparameters of a Bi-LSTM-CNN model using Bayesian Optimization. Initially, Bayesian Optimization is initialized, followed by defining the hyperparameter search space. An objective function assesses the model\u0026apos;s performance based on specified metrics. The algorithm iterates through a set number of iterations, selecting hyperparameters, training the model, evaluating its performance, and updating the Bayesian Optimization process. The optimal hyperparameters are obtained once the iterations are completed, and the final model is trained using these parameters. Finally, the performance of the trained model is evaluated, and the results are outputted. This approach automates the hyperparameter tuning process, allowing for efficient exploration of the hyperparameter space to find configurations that optimize the model\u0026apos;s performance. The overall process can be expressed in the following steps.\u003c/p\u003e\n \u003cdiv\u003e\n \u003cp\u003eStep 1: Initialize the Bayesian optimization algorithm.\u003c/p\u003e\n \u003cp\u003eStep 2: Define the range of hyperparameters to be optimized.\u003c/p\u003e\n \u003cp\u003eStep 3: Define the performance metric (e.g., accuracy, F1-score) used to evaluate the model.\u003c/p\u003e\n \u003cp\u003eStep 4: Bayesian Optimization Loop\u003c/p\u003e\n \u003c/div\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eUse the Bayesian optimization algorithm to select the next set of hyperparameters to evaluate.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eTrain the BiLSTM-CNN model using the selected hyperparameters.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eEvaluate the model\u0026apos;s performance on a validation set using the defined objective function.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eUpdate the Bayesian optimization algorithm with the hyperparameters and corresponding model performance.\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003cdiv\u003e\n \u003cp\u003eStep 5: Once the optimization loop is complete, obtain the optimal set of hyperparameters.\u003c/p\u003e\n \u003cp\u003eStep 6: Train the final BiLSTM-CNN model using the optimal hyperparameters obtained from Bayesian optimization.\u003c/p\u003e\n \u003cp\u003eStep 7: Evaluate the performance of the final model on a separate test set.\u003c/p\u003e\n \u003cp\u003eStep 8: Output the final model performance and any other relevant results.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e"},{"header":"3. Results and Observations","content":"\u003cp\u003eThis section discusses the study's dataset and the evaluation parameters for validating model performance. This is followed by an extensive analysis using diverse machine learning and deep learning algorithms. The overall performance is compared with some previous research works to make observations.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Dataset\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe dataset \"Satellites and Debris in Earth's Orbit,\" sourced from Kaggle, was obtained through an API provided by space-track.org. Comprising 14,372 rows and 40 columns, it encompasses attributes such as object name, ID, country code, and launch date. The target variable, object type, categorizes entities into payload, debris, rocket body, and TBA (to be announced). The data will be partitioned into 80% training and 20% test datasets for our experimental analysis. A significant imbalance in the dataset was noticed during the initial exploratory analysis. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e illustrates the overall distribution of the data. Notably, 58.7% of the data is labeled as debris, while 34.4% corresponds to the payload. Rocket body data accounts for 5.2%, with the remaining unnamed data. To mitigate bias, the study applies sampling techniques to balance the data before conducting a comprehensive analysis. Similarly, the RCS size can be classified as small, large, and medium.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Evaluation Parameters\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe following evaluation parameters have been considered for analyzing the model performance\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eAccuracy: Accuracy represents the proportion of correctly classified instances among the total number of instances. It gives an overall measure of how well the model performs across all classes. A higher accuracy indicates that the model is making fewer mistakes in its predictions.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ePrecision: Precision measures the proportion of true positive predictions (correctly predicted positive instances) out of all instances predicted as positive. It indicates the accuracy of positive predictions and is particularly relevant when the cost of false positives is high. A high precision means that the model is good at avoiding false positives.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eRecall: Recall, also known as sensitivity or true positive rate, measures the proportion of true positive predictions out of all actual positive instances. It indicates the model's ability to identify all relevant instances and is crucial when the cost of false negatives is high. A high recall means that the model is good at capturing positive instances.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eF-1 score: The F-1 score is the harmonic mean of precision and recall. It provides a single score that balances both precision and recall. It's useful when an uneven class distribution or false positives and negatives have different costs. A higher F-1 score indicates better overall performance, balancing precision and recall.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eTime to run (in seconds): Time to run refers to the time it takes for the model to train on the given data and make predictions. Considering this parameter, especially for large datasets or complex models, is essential as it impacts the model's scalability and efficiency. Minimizing the time to run is often desirable, but balancing it with achieving satisfactory performance metrics is crucial.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Experimental Analysis\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eA diverse set of algorithms, including classical machine learning techniques, deep learning techniques, and ensembles, has been considered for the experimental analysis. The study deploys logistic regression (LR), K-Nearest Neighbors (KNN), Gaussian Naive Bayes (GNB), Decision Trees (DT), Random Forests (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), Light Gradient Boosting (LGB), Support Vector Machines (SVM), Artificial Neural Networks (ANN), Feed Forward Neural Networks (FFNN), Long Short Term Memory (LSTM), Convolutional Neural Networks (CNN), and CNN-LSTM models on the dataset for classification. The performance of the proposed method, bi-LSTM-CNN with Bayesian Optimization, is compared against the diverse set of algorithms using evaluation metrics. The first phase of our study involves deploying classification models for identifying object types in the dataset, while the second phase involves identifying the RCS size (small, medium, large).\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of machine learning models for identifying Object Type\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorithm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF-1 Score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTime (s)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLR\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8744\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8109\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8246\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.332\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eKNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8515\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8278\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.870\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGNB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8492\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8384\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDT\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9132\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8422\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.114\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRF\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9439\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8567\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8897\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.782\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9287\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8018\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9208\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.654\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eXGB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9436\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9118\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.161\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLGB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8505\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9046\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e3.112\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSVM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9328\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8646\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9108\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e8.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eANN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9208\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8104\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8755\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6.038\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFFNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9451\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9036\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9228\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e14.732\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLSTM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9344\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8877\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e12.343\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9678\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9412\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9478\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e20.861\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCNN-LSTM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9604\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9238\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9558\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e28.446\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBi-LSTM- CNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9864\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9123\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9766\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e24.889\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBi-LSTM- CNN with Bayesian Optimization\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9916\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9604\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e47.662\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the performance evaluation of a comprehensive array of algorithms utilized in the analysis. The evaluation criteria include accuracy, precision, recall, F1 score, and the duration taken for model deployment. Across most algorithms, satisfactory performance is observed for multiclass classification. Logistic regression, K-nearest neighbor, and Gaussian naive Bayes exhibit relatively lower accuracy levels at 87%, 85%, and 84%, respectively. However, they boast shorter training times of 1.332s, 0.87s, and 1.001s, respectively. Conversely, neural networks like CNN and CNN-LSTM demonstrate excellent performance with accuracy rates of 96%, accompanied by longer training times of 20.86 seconds and 28.446 seconds, respectively. Bi-LSTM-CNN surpasses others in accuracy with 98% but requires a longer training duration of 24.889s. Notably, our proposed model, Bi-LSTM-CNN with Bayesian Optimization (Bi-LSTM-CNN-BO), achieves the highest accuracy of 99.16%, albeit with a training time of 47.66s. The rise in training time may be attributed to several reasons. BiLSTM-CNN models combine the complexities of both Bidirectional Long Short-Term Memory (BiLSTM) and Convolutional Neural Networks (CNNs), requiring substantial computational resources to train. Bayesian optimization explores a wide range of hyperparameters to find the optimal configuration for the model, which involves training and evaluating the model multiple times, increasing the overall training time. Moreover, Bayesian optimization may require many iterations before converging to the optimal set of hyperparameters, prolonging the training process as it iteratively refines the model configuration. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e depict the performance evaluation with respect to accuracy and time (seconds) for all the deployed algorithms.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of machine learning models for identifying RCS Size\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorithm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF-1 Score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTime (s)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLR\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9104\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9040\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.414\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eKNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9122\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8566\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.668\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGNB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8989\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8678\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8944\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.003\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDT\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9111\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8899\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e9.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e1.182\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRF\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9389\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7584\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7784\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2.324\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8333\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8541\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2.882\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eXGB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9123\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8717\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2.006\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLGB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9189\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8678\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8802\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2.267\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSVM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9465\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8686\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8993\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e7.067\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eANN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9499\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9129\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9118\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e5.114\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFFNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9312\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8504\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8894\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e11.998\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLSTM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9552\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8436\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8902\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e17.614\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9799\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8211\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8894\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e23.889\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCNN-LSTM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9804\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9004\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9122\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e38.775\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBi-LSTM- CNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9899\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9234\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9403\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e45.222\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBi-LSTM- CNN with Bayesian Optimization\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9998\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.9666\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9438\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e53.023\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrates the performance evaluation of a comprehensive array of algorithms utilized in the analysis. The evaluation criteria include accuracy, precision, recall, F1 score, and the duration taken for model deployment. Across most algorithms, satisfactory performance is observed for multiclass classification. Logistic regression, K-nearest neighbor, and Gaussian naive Bayes exhibit relatively lower accuracy levels at 93%, 91%, and 89%, respectively. However, they boast shorter training times of 0.414s, 0.668s, and 1.003s, respectively. Conversely, neural networks like CNN and CNN-LSTM demonstrate excellent performance with accuracy rates of 97.99% and 98.04%, accompanied by longer training times of 23.889s and 38.775s, respectively. Bi-LSTM-CNN surpasses others in accuracy with 98.99% but requires a longer training duration of 45.222s. Notably, our proposed model, Bi-LSTM-CNN with Bayesian Optimization (Bi-LSTM-CNN-BO), achieves the highest accuracy of 99.98%, albeit with a training time of 53.023 seconds. Figure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e depict the performance evaluation with respect to accuracy and time (seconds) for all the deployed algorithms.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Comparative Analysis\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eIn this segment, a comparative examination between our proposed approach and several existing methods for identifying space debris has been conducted. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e outlines the results of this comparative analysis.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparative Analysis of proposed work with relevant works\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor and Year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eProposed Work\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMethodology/ Parameters\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eResults\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJordan et al, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSpace Debris Removal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePSO-based method to estimate inertia parameters for uncooperative satellites\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eThe proposed method exhibits less than 1% error\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRyan et al, 2023]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOrbital debris impact risk assessments\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML algorithms like ANN, SVM, XGBoost, etc\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost shows the best performance with an accuracy 97.4%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGuo et al, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClustering spectral polarization data from space debris\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHybrid fuzzy C-means (FCM) algorithm model incorporating hierarchical agglomerative clustering (HAC)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAccuracy of 96.92%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQashoa and Lee, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClassifying low-orbit space objects\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAlgorithms like SVM, LSTM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLSTM exhibits an accuracy of 92%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhao et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTask Allocation for Space Debris Removal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eImproved Particle Swarm Optimization (PSO) Algorithm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eImproved PSO is 22.8% faster than traditional PSO\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eOur Proposed Work\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMonitoring and Classifying Space Debris\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003eExtensive ML algorithms, proposed Bi-LSTM-CNN, and Bayesian Optimization\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eThe proposed method archives accuracy of 99.16% and 99.98%\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Observations\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eSeveral key points have been observed based on the experimental and comparative analysis.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003ePrevious research in space exploration leveraging AI has been extensive but focused primarily on specific methodologies.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThis study conducts a thorough analysis employing diverse algorithms, contributing to the field of space debris analysis.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eMachine learning algorithms (classical, ensembles, neural networks) are assessed based on key performance metrics such as precision, accuracy, recall, F1 score, and deployment time.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eA robust Bi-LSTM-CNN model optimized with Bayesian techniques emerges as the top performer across multiple analyses, showcasing its efficacy in space debris identification and classification.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe initial analysis involves classifying debris based on object type, including debris, payload, and rocket bodies. In contrast, the subsequent analysis focuses on identifying RCS size, a critical factor for understanding debris characteristics and potential hazards.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eComparative analysis reveals that this approach achieves satisfactory accuracy compared to previous works, highlighting its effectiveness in identifying and monitoring space debris.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Limitations\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eSome limitations of the study are as follows:\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eThe study utilized a specific dataset for analysis, which may not fully represent the entire spectrum of space debris characteristics and behaviors. This could affect the generalizability of the findings.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThe study assumes the quality and reliability of the dataset used for analysis. However, data inconsistencies, missing values, or errors could potentially introduce biases and affect the robustness of the results.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThe complexity of the Bi-LSTM-CNN model with Bayesian optimization may lead to challenges in interpretability and scalability. Additionally, the computational resources required to train and deploy this model may limit its practical applicability in certain contexts.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eExternal factors such as changes in space policies, advancements in technology, or unforeseen events could influence the dynamics of space debris and render the study findings outdated or less relevant over time.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Conclusion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe comprehensive analysis conducted in this study sheds light on the intricate task of space debris identification and classification, presenting valuable insights and potential avenues for further research. Through the utilization of diverse machine learning algorithms, including logistic regression, K-nearest neighbors, Gaussian naive Bayes, and neural networks like CNN and CNN-LSTM, the performance of various models in classifying space debris types and predicting Radar Cross Section (RCS) sizes has been examined. Additionally, introducing a robust Bi-LSTM-CNN model optimized with Bayesian techniques showcased remarkable accuracy in both classification tasks, albeit with a longer training time attributed to the model's complexity. The findings underscore the importance of accurately identifying and monitoring space debris, which poses significant risks to space missions and satellite operations. Despite the satisfactory performance achieved by the proposed model, it is crucial to acknowledge the study's limitations. The reliance on a specific dataset may limit the generalizability of the results, and potential data quality issues could introduce biases.\u003c/p\u003e \u003cp\u003eFuture research endeavors should address these limitations by incorporating larger and more diverse datasets, ensuring data quality and reliability, and exploring advanced techniques to enhance model performance and efficiency. Additionally, efforts should be directed toward integrating real-time monitoring systems and predictive analytics to mitigate the growing threat of space debris and safeguard space infrastructure. In conclusion, while the study marks a significant step towards advancing space debris analysis, it also highlights the need for continuous innovation and collaboration within the scientific community to address the challenges posed by space debris and ensure the sustainability of space exploration endeavors.\u003c/p\u003e \u003c/div\u003e "},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eI.P. wrote the article, performed analysis, proofreadsingle author contribution\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eFunding Declaration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research project received no external funding. The author declares that they have not received any financial support or funding from any organization, institution, or individual for this research.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eData Availability Declaration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets cited in this manuscript are publicly available and can be accessed from the original sources as referenced in the text. No additional data were generated or analyzed during the current study\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eCompeting Interest declaration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests regarding the publication of this manuscript\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSvotina, V. V., \u0026amp; Cherkasova, М. V. (2023). Space debris removal\u0026ndash;Review of technologies and techniques. Flexible or virtual connection between space debris and service spacecraft. Acta Astronautica, 204, 840-853.\u003c/li\u003e\n\u003cli\u003eLedkov, A. S., \u0026amp; Aslanov, V. S. (2023). Active space debris removal by ion multi-beam shepherd spacecraft. Acta Astronautica, 205, 247-257.\u003c/li\u003e\n\u003cli\u003eWang, C., Chen, D., Liao, W., \u0026amp; Liang, Z. (2023). Autonomous obstacle avoidance strategies in the mission of large space debris removal using potential function. Advances in Space Research, 72(7), 2860-2873.\u003c/li\u003e\n\u003cli\u003eViavattene, G., Devereux, E., Snelling, D., Payne, N., Wokes, S., \u0026amp; Ceriotti, M. (2022). Design of multiple space debris removal missions using machine learning. Acta Astronautica, 193, 277-286.\u003c/li\u003e\n\u003cli\u003eWu, Y., Peng, K. L., Yao, Y., \u0026amp; Guo, Y. (2024). Sustainable Space Travel: What Can We Do in Education from Economic and Environmental Perspectives?. Sustainability, 16(2), 684.\u003c/li\u003e\n\u003cli\u003eDeng, S., Su, J., Zhu, Y., Yu, Y., \u0026amp; Xiao, C. (2024). Forecasting carbon price trends based on an interpretable light gradient boosting machine and Bayesian optimization. Expert Systems with Applications, 242, 122502.\u003c/li\u003e\n\u003cli\u003eKhan, M. A., Arshad, H., Khan, W. Z., Alhaisoni, M., Tariq, U., Hussein, H. S., ... \u0026amp; Elashry, A. (2023). HGRBOL2: human gait recognition for biometric application using Bayesian optimization and extreme learning machine. Future Generation Computer Systems, 143, 337-348.\u003c/li\u003e\n\u003cli\u003eJordan, J., Posada, D., Zuehlke, D., Nocerino, A., Fontdegloria, P., John, S., ... \u0026amp; Henderson, T. (2023, March). Estimation of Uncooperative Satellite Inertia Parameters for Space Debris Removal Using Particle Swarm Optimization. In 2023 IEEE Aerospace Conference (pp. 1-13). IEEE.\u003c/li\u003e\n\u003cli\u003eRyan, S., Sushma, N. M., Le, H., Kumar, A. A., Berk, J., Nguyen, T. M., ... \u0026amp; Venkatesh, S. (2023). The application of machine learning in micrometeoroid and orbital debris impact protection and risk assessment for spacecraft. International Journal of Impact Engineering, 181, 104727.\u003c/li\u003e\n\u003cli\u003eWu, D., \u0026amp; Rosengren, A. J. (2023). An investigation on space debris of unknown origin using proper elements and neural networks. Celestial Mechanics and Dynamical Astronomy, 135(4), 44.\u003c/li\u003e\n\u003cli\u003eXu, Y., Liu, X., He, R., Zhu, Y., Zuo, Y., \u0026amp; He, L. (2023). Active Debris Removal Mission Planning Method Based on Machine Learning. Mathematics, 11(6), 1419.\u003c/li\u003e\n\u003cli\u003eCimmino, N., Opromolla, R., \u0026amp; Fasano, G. (2023). Machine learning-based approach for ballistic coefficient estimation of resident space objects in LEO. Advances in Space Research, 71(12), 5007-5025.\u003c/li\u003e\n\u003cli\u003eSalmaso, F., Trisolini, M., \u0026amp; Colombo, C. (2023). A machine learning and feature engineering approach for the prediction of the uncontrolled re-entry of space objects. Aerospace, 10(3), 297.\u003c/li\u003e\n\u003cli\u003eGuo, F., Zhu, J., Huang, L., Li, H., Deng, J., Jiang, H., \u0026amp; Hou, X. (2023). Enhancing Spatial Debris Material Classifying through a Hierarchical Clustering-Fuzzy C-Means Integration Approach. Applied Sciences, 13(8), 4754.\u003c/li\u003e\n\u003cli\u003eSingh, G., Kumar, V., Buduru, A. B., \u0026amp; Biswas, S. K. (2024). Tracking an untracked space debris after an inelastic collision using physics informed neural network. Scientific Reports, 14(1), 3350.\u003c/li\u003e\n\u003cli\u003eZhang, H., Zhang, Y., Feng, Q., \u0026amp; Zhang, K. (2024). Review of Machine-Learning Approaches for Object and Component Detection in Space Electro-optical Satellites. International Journal of Aeronautical and Space Sciences, 25(1), 277-292.\u003c/li\u003e\n\u003cli\u003ePensado, E. A., de Santos, L. M. G., Sanjurjo-Rivo, M., \u0026amp; Jorge, H. G. (2023). Deep Learning Based Target Pose Estimation Using LiDAR Measurements in Active Debris Removal Operations. IEEE Transactions on Aerospace and Electronic Systems.\u003c/li\u003e\n\u003cli\u003eQashoa, R., \u0026amp; Lee, R. (2023). Classification of Low Earth Orbit (LEO) Resident Space Objects\u0026rsquo;(RSO) Light Curves Using a Support Vector Machine (SVM) and Long Short-Term Memory (LSTM). Sensors, 23(14), 6539.\u003c/li\u003e\n\u003cli\u003eZhao, Y., Wu, P., Yu, H., \u0026amp; Tong, Y. (2023, October). Task Allocation for Space Debris Removal based on Improved Particle Swarm Optimization Algorithm. In 2023 International Conference on Cyber-Physical Social Intelligence (ICCSI) (pp. 399-404). IEEE.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Bayesian Optimization, Deep Learning, Monitoring System, Space Debris, Bi-LSTM-CNN","lastPublishedDoi":"10.21203/rs.3.rs-3968729/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3968729/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMonitoring space debris is vital for ensuring the safety of space missions and satellite operations amid the increasing number of satellites and spacecraft in orbit. The study addresses this challenge by proposing a novel approach based on a hybrid Bi-LSTM-CNN architecture optimized using Bayesian Optimization. Through extensive analysis utilizing machine learning and deep learning techniques, the study develops a robust space debris detection system capable of classifying both the object type and Radar Cross Section (RCS) size. The proposed method outperforms existing approaches by demonstrating superior performance across multiple evaluation metrics, including accuracy, precision, recall, and F1 score. Moreover, the study considers the practical aspect of training time, ensuring efficiency in real-time applications. Empirical validation on real-world datasets confirms the effectiveness and efficiency of the hybrid model in accurately detecting and predicting space debris types. Overall, this research significantly advances space debris monitoring capabilities, mitigating risks associated with space exploration and satellite operations, and offers comprehensive insights into potential hazards and optimizing mitigation strategies.\u003c/p\u003e","manuscriptTitle":"Enhanced Space Debris detection and monitoring using a hybrid Bi-LSTM- CNN and Bayesian Optimization","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-05 18:55:31","doi":"10.21203/rs.3.rs-3968729/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"73012d16-e0b3-4831-b053-381898a1dc04","owner":[],"postedDate":"March 5th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-03-06T13:38:25+00:00","versionOfRecord":[],"versionCreatedAt":"2024-03-05 18:55:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3968729","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3968729","identity":"rs-3968729","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}