{"paper_id":"4cf675f0-d741-4f25-8955-cf05e5e4f51d","body_text":"Towards Smart Monitoring of Invasive Aquatic Plants: A Comparative Evaluation of YOLOv11 and Faster R-CNN for Water Hyacinth Detection | 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 Towards Smart Monitoring of Invasive Aquatic Plants: A Comparative Evaluation of YOLOv11 and Faster R-CNN for Water Hyacinth Detection Richard Mkechera, Sonika Dahiya This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7323485/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 In recent years, the proliferation of water hyacinths across various freshwater bodies has become an escalating environmental concern due to its adverse effects on agricultural productivity, aquatic ecosystems, and public health. As a highly invasive aquatic weed, water hyacinth rapidly spreads, obstructing irrigation canals, drainage systems, and fish farms, depleting oxygen levels, and degrading water quality, thereby directly challenging the objectives of Sustainable Development Goal (SDG) 6, which calls for improved water quality and the protection of water-related ecosystems. Early detection is essential for controlling its spread, yet accurately distinguishing water hyacinths from other aquatic vegetation remains technically difficult. To address this, this study evaluates the effectiveness of deep learning models in the rapid and accurate detection of water hyacinths. Specifically, we conduct a comparative analysis of YOLOv11n and Faster R-CNN using two datasets: a custom WaterHyacinth dataset and the Kaggle Water Hyacinth dataset. The models are assessed based on accuracy, inference speed, model size, and mean average precision (mAP). Our findings indicate that YOLOv11n outperforms Faster R-CNN, achieving 96.3% and 93.8% accuracy, on the respective datasets. Furthermore, YOLOv11n’s lightweight architecture and faster inference speed make it more suitable for real-time applications, highlighting its potential as a practical tool for the early detection and management of water hyacinth, thus contributing to SDG 6’s mission of ensuring clean water and healthy freshwater ecosystems. Water Hyacinth YOLOv11 Faster R-CNN Inference speed detection invasive aquatic plants Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Water resource conservation is a critical component of environmental sustainability, essential for preserving the biodiversity and ecological balance of aquatic ecosystems such as wetlands, lakes, and rivers. It includes the protection of both water quality and quantity, ensuring the long-term viability of freshwater systems. This aligns directly with the United Nations (UN) Sustainable Development Goal 6 (SDG 6), which emphasizes the need to ensure the availability and sustainable management of water and sanitation for all. Achieving this goal requires integrated efforts to monitor, manage, and safeguard freshwater ecosystems from degradation and invasive threats. These water bodies serve as a vital resource for drinking water, fisheries, agriculture, transportation, and habitat for aquatic organisms. Moreover, increasing anthropogenic pressure, like pollution, industrial discharge, invasive species like water hyacinth, and agricultural runoff have seriously degraded these ecosystems, threatening their ecological balance and the services they provide (Erftemeijer & Robin Lewis, 2006 ). Conservation efforts involve a combination monitoring of scientific monitoring, policy-making, community engagement, and application of modern technologies such as deep learning, remote sensing, and machine learning for effective management. By adopting sustainable practices and proactive restoration measures, water body conservation ensures the long-term availability and health of aquatic organisms, which are essential for human well-being and environmental stability(Merlin, 2011 ). Aquatic weeds, generally invasive species such as water hyacinth (Eichhornia crassipes) have caused significant challenges in freshwater ecosystems, and ecological and economic challenges (Emary et al., 2014 ). Fast-growing plants can rapidly cover the surface of lakes, reservoirs, and rivers, block sunlight, and reduce oxygen levels in the water (“Water Hyacinth (Eichhornia Crassipes) Biology and Its Impacts on Ecosystem, Biodiversity, Economy and Human Well-Being,” 2019). Their dense mats hinder water flow, affect fishing activities, and clog irrigation canals together with hydroelectric systems, leading to an increase in maintenance costs and poor productivity (Onyari et al., 2024 ). Also, aquatic weeds can become breeding grounds for diseases like malaria, which leads to the contribution of public health risks. However, proliferation normally starts from nutrient-rich water bodies by untreated water and agricultural runoff. Controlling the spread of water hyacinth requires integrated management strategies, and advanced monitoring and detection techniques, such as the application of deep learning and machine learning techniques which can enhance the elimination of these weeds. Object detection is the technique that locates and identifies objects within images or a video. Objects are classified based on their features such as shape, size, and color within a defined box known as a bounding box (Manojkumar et al., 2023 ). However, in water hyacinth, object detection is applicable to automatically find and mark the presence of these invasive weeds in water bodies using ground-level or aerial images. By training deep learning models on labeled datasets of water hyacinth. These models can accurately detect and locate water hyacinths in both simple and complex environments. In recent years deep learning models have emerged as some of the most effective algorithms for object detection, significantly enhancing the accuracy of computer vision-based systems. To improve the detection of water hyacinths, this study explores the role of deep learning models in advancing the identification and monitoring of aquatic weeds (Hao et al., 2024 ). This work makes the following key contributions to the field of environmental monitoring: This research work offers valuable insights for selecting suitable models for water hyacinth detection. This research work contributes a meticulously annotated dataset formatted for both COCO and YOLO standards, facilitating future studies in water hyacinth detection. This research uses a dual-dataset approach to enhance the diversity and reliability of the evaluation framework. 2. Deep Learning Models for Detecting Water Hyacinth To support the automated detection of water hyacinths in remote sensing imagery, this study employs two state-of-the-art object detection models: YOLOv11 (Khanam & Hussain, 2024 ) and Faster R-CNN (Ren et al., n.d.). These architectures offer complementary strengths in speed and precision, respectively, which are critical for balancing real-time monitoring with detection accuracy in aquatic ecosystems. Object detection techniques have spearheaded the challenges of identification and localization of an image in real-world scenarios. The evolution of object detection solved a lot of the computer vision problems, some of the object detection which brought hope in the field of computer vision are Single Short Multibox Detector (SSD) (Liu et al., 2015 ), Region Proposal-Convolution Neural Network (R-CNN)(Girshick et al., 2013 ), YOLO (You Only Look Once) (Redmon et al., 2015 ). This paper chooses to work with YOLOv11 and faster R-CNN, to contribute to the strategic plans for the removal of water hyacinth. Advancement of the detection techniques leads to the solution of various problems existing in the community due to the high detection applicability on a particular object, although the detection of these models is highly affected by low resolution, occlusion, and complex environmental conditions (Islam & Islam, n.d.). Deep learning models have been grouped into two major groups (1) Two-stage detector which at first identifies potential object location, this is done before object classification and bounding box regression, this architecture allows to detection of an object preciously but it's more resource intensive making it more complex for real-time application (2) a one stage detector processes the entire image in a single pass, allowing improved inference speed and lower memory usage while maintaining a competitive accuracy (Nikita Singhal & Lalji Prasad, 2023). Understanding the performance and applicability of deep learning models toward the implementation of the embedded systems needs a detailed comparative analysis of different deep learning models in various environmental scenarios (Dakari Aboyomi & Daniel, 2023a ), This makes it highly suitable for applications requiring real-time detection. This work highlights that model selection should comply with the specific need of the task, whether it is rapid real-time detection or high precision, this demonstrates that no single model is universally superior but rather depends on the performance of optimal requirements. 2.1 Two-stage detector The two-stage detection architecture involves a sequential stage where the model generates a series of region proposals that represent areas of the image that are likely to present the object to be detected, classifies each region, and refines the corresponding bounding box (Asudani et al., 2024 ). R-CNN (Girshick et al., 2013 ) introduced a two-stage detector framework by applying selective search to propose regions, which were passed through a Convolutional Neural Network (CNN) to extract features of the object to be detected. Features were extracted using a Support Vector Machine (SVM) (Cortes et al., 1995 ), and bounding box regressors were used to refine the locations. This model recorded a high accuracy compared to previous traditional models, but it was computationally expensive due to high computations on the overlapping regions. Since the introduction of R-CNN, there have been variants of R-CNN which were used to address the challenges faced by the R-CNN. Fast R-CNN (Girshick, 2015 ), was proposed to address the challenges of R-CNN by integrating region proposal and feature extraction in a single Convolution Neural Network (CNN) forward pass, in this case, the entire image is processed by CNN only once; while generating a feature map and regions of interests are extracted from this region using pooling layer. This improvement of R-CNN reduced computation load and improved training speed to a large extent. Upon improving the efficiency of the Fast R-CNN, a new variant of R-CNN was introduced called Faster R-CNN (Ren et al., n.d.), this was more optimized when the Regional Proposal Network (RPN) was introduced, which shares a convolution feature with detection network and predicts region proposal directs by omitting the need for external proposal methods such as selective search. Faster R-CNN focuses more effectively on the complex scenario especially when the object varies in size and shape. Table 1 below shows architectural differences and improvements of R-CNN. Table 1 Two-Stage Detector: R-CNN Ref Model Religion Proposal Detection Network (Girshick et al., 2013 ) R-CNN Selective search SVM (Girshick, 2015 ) FAST RCNN Selective search SoftMax + Bounding box regressor (Ren et al., n.d.) FASTER RCNN Region Proposal Network (RPN) SoftMax + Bounding box regressor 2.2 One-stage detector With the improvement of the two-stage object detector from R-CNN to faster R-CNN, object detection accuracy remained to be a very huge challenge. To solve all these challenges a stage detector was introduced. YOLO (You Only Look Once) (Redmon et al., 2015 ), revolutionized real-time object detection by considering direct regression of the bounding boxes and class probabilities in a single pass through a neural network. YOLO reduces computational overhead while maintaining competitive accuracy, making them suitable for real-time applications. The invention of the YOLO algorithm marked the beginning of the journey of other YOLO versions, from YOLOv2 to YOLOv11, as shown in Table 2 . YOLOv1 (Redmon et al., 2015 ), introduced the journey of a single-stage object detector that defined it as a stage regression problem while maintaining high inference speed but failing to detect objects of small size and overlapping objects. The limitation of YOLOv1 was improved in YOLOv2 (YOLO9000)(Redmon & Farhadi, 2016 ), in this version they fixed the problem of detecting objects of small size by introducing anchor boxes and batch normalization. YOLOv3(Redmon & Farhadi, 2018 ) greatly improved accuracy compared to YOLOv1 and YOLOv2, by introducing a backbone network which is Darknet-53 and multi-scale detection. To enhance feature extraction YOLOv4 (Bochkovskiy et al., 2020 ), was introduced with the addition of features like CSPDarknet53, PANeT for feature aggregation, and Mosaic augmentation which allows the model to see the object into multiple contexts and object scales within a single forward pass. To improve the performance of YOLOv4, YOLOv5(Jocher G et al, 2020 ) was introduced and was greatly improved in terms of performance and accuracy where auto-learning bounding box anchors were introduced to improve accuracy and reduce the necessity of manual tunning, also inference speed was increased and training time was reduced compared to YOLOv4. Modularity and scalability were the key improvements of YOLOv5 where a number of multiple model sizes such as YOLOv5s, YOLOv5m, and YOLOv5l were introduced. YOLOX (Ge et al., 2021 ), This version improved model generalization and simplified the model by introducing anchor-free design which makes it different from traditional anchor-based detection used in different versions of YOLO. Additionally, YOLOX applied a decoupled head for regression and classification, addressing the challenges observed in various versions of YOLO. Therefore, the improvement made in this model improved the accuracy of the YOLO model to a very great extent. Industrial applications and model deployment were very big questions due to the model size and performance in a real-time scenario, to improve model deployment capabilities and performance of a model YOLOv6(Li et al., 2022 ) was introduced at which featured an anchor-free design, decoupled head architecture and improved training strategy. This model outperformed the previous version of YOLO due to high accuracy and precision. YOLOv7 (C.-Y. Wang et al., 2023 ), introduced a re-parameterized convolution layer and extended efficiency layer aggregation efficiency network (E-ELAN). This model set up a new record for model accuracy and speed, therefore it became more useful for real-time object detection. YOLOv8(Ultralytics, 2023 ), this version continues with the improvement of backbone redesigning, decoupled head, and segmentation tasks. This model is too flexible which makes it ideal for practical deployment across multiple domains. In YOLOv9 (C.-Y. Wang et al., 2024 ), Generalized Efficient Layer Aggregation Network (GELAN) and programmable gradient information (PGI) were introduced. These significantly improved feature reuse and inference speed, especially for edge device problems. All versions of YOLO had a limitation on computation cost and inference speed, but in YOLOv10 (A. Wang et al., 2024 ), they introduced a dual assignment for non-maximum suppression (NMS)-free training which resulted in low inference latency. To keep working on model size for easy deployment, YOLOv11(Khanam & Hussain, 2024 ) was introduced, where a unified neural network architecture that handles both bounding box regression and classification task was introduced, and in backbone network a newly cross-stage partial with spatial attention which improved accuracy due to the effectiveness of feature maps. Table 2 One-Stage Detector: YOLO Ref Model Improvement Detection Network (Redmon et al., 2015 ) YOLOv1 Single Stage for classification and Regression problem CNN (YOLO) (Redmon & Farhadi, 2016 ) YOLOv2 (YOLO9000) Anchor boxes, Batch normalization, and High resolution Darknet-19 (Redmon & Farhadi, 2018 ) YOLOv3 Multi-Scale prediction and residue blocks Darknet-53 (Bochkovskiy et al., 2020 ) YOLOv4 Bag of freebies CSPDarknet-53 (Jocher G et al, 2020 ) YOLOv5 Model scaling and Auto-learning bounding box New CSPDarknet-53 (Ge et al., 2021 ) YOOX Ancho- free and decoupled head Modified CSPNet (Li et al., 2022 ) YOLOv6 Anchor-free support and EfficientRep (C.-Y. Wang et al., 2023 ) YOLOv7 ELAN and extended reparameterization ELAN backbone (Ultralytics, 2023 ) YOLOv8 Advanced argumentation and new decoupled head Modified CSPnet (C.-Y. Wang et al., 2024 ) YOLOv9 Programmable gradient Information which improved accuracy GELAN (A. Wang et al., 2024 ) YOLOv10 Lightweight design and enhanced real-time performance Optimized CSPNet (Khanam & Hussain, 2024 ) YOLOv11 Speed and Memory Stage Partial with Spatial Attention (C2PSA) 3. Proposed method This study adopts an experimental design (Dakari Aboyomi & Daniel, 2023b ) to examine the performance of FASTER R-CNN(Ren et al., n.d.) and YOLOv11(Khanam & Hussain, 2024 ) for water hyacinth detection. Two annotated datasets; the WaterHyacinth datasets (Kabir et al., 2024 ) and the Kaggle Water Hyacinth dataset (kaggle, n.d.). All datasets are preprocessed through proposed stages and trained through the training pipeline of each model with performance assessed across accuracy, speed, and computational efficiency metrics. The methodology emphasizes reproducibility, with detailed information on filtering techniques, augmentation strategies, and model training processes. The methodology is well illustrated in Fig. 1 . The proposed pipeline aims to compare the performance of two object detection models; Faster R-CNN and YOLOv11 for the task of Water Hyacinth detection, the pipeline consists of multiple structured phases. 3.1 Dataset acquisition This work uses two distinct datasets; the WaterHyacinth dataset (Kabir et al., 2024 ) and the Kaggle Water Hyacinth dataset (kaggle, n.d.), which comprises varied resolution, lighting, and scene complexity for robust models' assurance as summarized on Table 3 to indicate the total number of images in each dataset. These datasets are publicly available datasets, due to their quality and the number of images in each dataset, we had to add some preprocessing techniques in our working flow to improve the quality of images and the number of images in each dataset. Table 3 Dataset Dataset category S.no Name of dataset Number of images 01 WaterHyacinth dataset (Kabir et al., 2024 ) 1000 02 Kaggle Water Hyacinth datasets (kaggle, n.d.) 584 3.2 Image Filtering This study uses image filtering, with the aim of improving image quality by reducing noise. This involves loading zipped image files, extracting images from zipped files, and applying spatial filtering techniques where median filtering has been applied due to its excellence in noise reduction capability (Gonzalez & Woods, 2018 ). Finally, it stores all the filtered images for subsequent steps, as illustrated in Fig. 1 . 3.3 Data argumentation As shown in Table 3 above, the number of images from both datasets is 584 for the Kaggle WaterHyacinth dataset and 1000 for WaterHyacinth. Dataset after doing a critical observation on both datasets, we realized that the size of data isn’t enough to train a model. To enhance model generalization rotation argumentation technique was employed, which is done by exposing the images to the different orientations of the same objects. The rotation parameter was carefully selected; slight rotation was made at an angle of 20°. A small angle is considered, because as the rotation angle increases data label is distorted, which may affect model generalization and accuracy of a model (Shorten & Khoshgoftaar, 2019 ). Following data augmentation, the total number of images increased to 2,000 and 1,732, respectively, as presented in Table 4 . Table 4 Dataset after argumentation Dataset category S.no Name of dataset Number of images 01 WaterHyacinth dataset (Kabir et al., 2024 ) 2000 02 Kaggle Water Hyacinth datasets (kaggle, n.d.) 1752 3.4 Data annotation and Splitting Since our study uses YOLOv11 and FASTER R-CNN. Therefore, there is a need to annotate images in two formats (1) YOLO format(Redmon et al., 2015 ) and (2) COCO format(Everingham et al., 2010 ). For YOLOv11, the annotation is well prepared using a specific YOLO format, which includes bounding box coordinates relative to the image dimension. For FASTER R-CNN, the annotation is structured in JSON COCO format enabling its compatibility and effectiveness in TensorFlow object detection API. The annotated data sets are then split into validation sets, Training sets, and testing sets in the ratio of 20:70:10 respectively, to ensure uniform label distribution across all sets (Everingham et al., 2010 ). 4. Implementation Object detection was implemented using YOLOv11 and FASTER R-CNN and the two different datasets: the Water Hyacinth dataset and the Kaggle Water Hyacinth dataset. Table 5 discusses the hardware and software requirements. Table 5 Experiment setup 01 Language Python 02 Environment GPU 03 Dataset WaterHyacinth dataset (Kabir et al., 2024 ) Kaggle Water Hyacinth datasets (kaggle, n.d.) 04 Preprocessing Image filtering 05 Augmentation angle 20° 06 Architecture YOLOv11 and FASTER R-CNN This section details the experiment setup and step-by-step methodology employed to train and evaluate two distinct datasets using both YOLOv11 and FASTER R-CNN architectures. This comparative analysis follows a systematic approach to ensure reproducibility of the results at the end of the experiment. Each architecture was trained independently on both datasets as indicated on both Fig. 2 and Fig. 4 respectively, and performance metrics such as loss curve, training time, model size, and inference speed were recorded. In addition to training our model, we systematically monitored the training loss throughout the entire training process. This enabled us to assess the model's learning behavior over time and identify any signs of overfitting or underfitting. The recorded training loss trends are presented in Fig. 3 and Fig. 5 respectively, offering insights into the model’s convergence and overall performance during training. 4.1 Analysis and comparison Result analysis is obtained by comparing the performance metrics of every architecture. Metrics considered during experimentation were inference speed, model size, and accuracy which gives a clear picture of which model is effective for water hyacinth detection. Since all datasets were split into test sets, then the test set was executed from both datasets, for the purpose of showing the effectiveness in detection and practical implication of each architecture. The analysis also aims to provide and suggest which model should be deployed to the embedded systems for easy and fast detection of water hyacinths in the aquatic environment. 4.2 Evaluation strategy The study has considered the inference speed as one of the metrics to evaluate model performance. In deep learning models inference speed measures how faster, the model processes the input data and produces the output (Bochkovskiy et al., 2020 ). The formula for the inference speed is: $$\\:\\text{I}\\text{n}\\text{f}\\text{e}\\text{r}\\text{e}\\text{n}\\text{c}\\text{e}\\:\\text{s}\\text{p}\\text{e}\\text{e}\\text{d}=\\frac{\\text{T}\\text{o}\\text{t}\\text{a}\\text{l}\\:\\text{n}\\text{u}\\text{m}\\text{b}\\text{e}\\text{r}\\:\\text{o}\\text{f}\\:\\text{i}\\text{m}\\text{a}\\text{g}\\text{e}\\text{s}\\:}{\\text{T}\\text{o}\\text{t}\\text{a}\\text{l}\\:\\text{r}\\text{e}\\text{f}\\text{e}\\text{r}\\text{e}\\text{n}\\text{c}\\text{e}\\:\\text{t}\\text{i}\\text{m}\\text{e}\\:\\left(\\text{i}\\text{n}\\:\\text{s}\\text{e}\\text{c}\\text{o}\\text{n}\\text{d}\\:\\right)}$$ 1 In addition to that, there are some metrics that have been considered in this study such as accuracy and precision. To ensure model prediction is robust, reliable, and good enough for the intended prediction, we must measure how accurate and precise the model is. $$\\:\\text{A}\\text{c}\\text{c}\\text{u}\\text{r}\\text{a}\\text{c}\\text{y}=\\frac{\\text{T}\\text{P}\\:+\\:\\text{T}\\text{N}\\:}{\\text{T}\\text{o}\\text{t}\\text{a}\\text{l}\\:\\text{s}\\text{a}\\text{m}\\text{p}\\text{l}\\text{e}\\text{s}}$$ 2 Where TP is referred to as True Positive and TN is referred to as True Negative. From (2) accuracy is the ratio of correctly predicted samples or instances to the total number of predictions. This metric always gives a quick overview to the researcher on the behavioral balance of the classes; with this regard, we have used this metric in this study as a comparison baseline metric. $$\\:\\text{P}\\text{r}\\text{e}\\text{c}\\text{i}\\text{s}\\text{i}\\text{o}\\text{n}=\\frac{\\text{T}\\text{P}\\:}{\\text{T}\\text{P}\\:+\\:\\text{F}\\text{P}}$$ 3 Where precision is defined as the ratio of True Positive and the total number of True Positive and False Positive, it answers the question, of how many samples are positive during the training, out of all predicted samples. It is more useful when understanding the correctness of the model. 5. Results and discussion This section presents and analyzes results obtained from the comparative evaluation of the Faster R-CNN and YOLOv11 models across multiple performance metrics, including accuracy, precision, mean Average Precision (mAP), and inference speed. The evaluation was conducted using two different datasets with varied characteristics to assess the models' effectiveness under different conditions. 5.1 Result by Dataset Table 6 WaterHyacinth dataset Metrics Faster R-CNN YOLOv11 Accuracy 0.85 0.963 mAP 0.72 0.977 Model size 294.43MB 16.21MB Training time 60min 24.18min Inference speed 4.8ms 4.5ms Table 7 Kaggle Water Hyacinth dataset Metrics Faster R-CNN YOLOv11 Accuracy 0.77 0.938 mAP 0.67 0.956 Model size 294MB 12.17MB Training time 55min 17.52min Inference speed 4.8ms 4.4ms The above results were obtained after training distinct datasets which were trained under the same settings. In this study, we have used multiple indicators (Accuracy, Mean Average Precision, model size, Training time, and inference speed) as the core metrics to evaluate Faster R-CNN and YOLOV11. The purpose of using the two different datasets is to increase the generalizability and clarity of findings on which model is useful for Water Hyacinth detection. As is seen in Table 6 and Table 7 respectively YOLOV11 significantly outperforms the performance of faster R-CNN in accuracy and computational efficiency over the two distinct datasets used in this study. The results explain how YOLOV11 is suitable for real-time Water Hyacinth detection applications, due to its low inference speed and small model size. The results suggest YOLOV11 can be useful particularly in mobile deployments and in edge-computing where speed and model size are seriously considered. Figure 6 , illustrates the comparative analysis of model accuracy between YOLOv11 and Faster R-CNN across two datasets. On both two sets of data YOLOv11, shows an outstanding performance in terms of accuracy. However, WaterHyacinth Dataset, YOLOv11 achieves an accuracy of 0.963, outperforming Faster R-CNN which achieves 0.85. For the Kaggle Water Hyacinth Datasets YOLOv11 attains 0.938 while Faster R-CNN scores a lower accuracy of 0.77. Figure 5 shows that YOLOv11 has stronger generalization capacities and is more effective in detecting and classifying Water Hyacinth instances across different water bodies. The steep upward trend from Faster R-CNN to YOLOv11 in both lines reinforces YOLOv11's advantage in terms of accuracy. Thus, the graph provides very clear empirical evidence supporting YOLOv11, as the best and most accurate model for the detection of Water Hyacinth in various aquatic environments compared to Faster R-CNN. The detection output shown in Fig. 7 , demonstrates the successful detection of water hyacinths across different visual scenarios, each input image contains a bounding box that accurately localizes Water Hyacinth clusters, regardless of their scale variation, background complexity, and lighting condition. All models demonstrate robustness to the environmental noises as they detect the target class reliably. This level of performance shows high spatial pressure and informs that the trained detector has effectively learned the distinguishing features of Water Hyacinth. 6. Conclusion and Future Work This research work introduces a comparative analysis of YOLOv11 and FASTER R-CNN for Water Hyacinth detection using two distinct datasets. The evaluation centered on the key performance metrics such as accuracy, inference speed, training time, and mean average precision (mAP). The findings of this research shown in Table 6 and Table 7 respectively, demonstrate that YOLOv11 consistently outperforms Faster R-CNN in terms of training efficiency and inference speed, recording significantly faster detection times and reduced time taken to train a model. However, YOLOv11 exhibited superior and higher mAP across both Kaggle Water Hyacinth and Water Hyacinth datasets, highlighting its capability for real-time and resource-efficient deployment. In addition, while faster R-CNN delivers competitive accuracy, it scores higher computation cost, high memory size, and slower inference speed, becoming less suitable for embedded applications. YOLOv11 emerges to be a more practical and scalable solution for real-time water hyacinth and other aquatic species detection and monitoring. The impact of the aquatic weeds in water sources is still very high and more challenging to many countries around the world, therefore this research work provides a third eye toward the process of eliminating aquatic weeds from water bodies. With these research findings, we suggest, to ensure the reliable elimination of water hyacinth in the future there should be an exploration of the integration of advanced data argumentation techniques such as Generative Adversarial Networks(GAN) and rotation-based transformation to increase the size of water hyacinth datasets which has become a very big challenge on implementing computer vision-oriented application to facilitate the elimination of the aquatic weeds, especially water hyacinth and to enhance model robustness under diverse environmental condition. In the future detection of water hyacinth may focus on extending the comparative evaluation of water hyacinth detection by incorporating lightweight transformer-based architectures, as well as exploring additional variants of the YOLOv11 model. Such models hold significant promise for enhancing performance, particularly in edge computing environments where computational resources are limited. The integration of real-time feedback mechanisms, linked with active learning frameworks, can further improve the model's adaptability to changing environmental conditions. This would facilitate continuous learning and ensure robust detection capabilities in dynamic and heterogeneous aquatic ecosystems. By adopting these advanced approaches, future systems could become more responsive, efficient, and scalable for widespread deployment in ecological monitoring and management applications. Declarations Author contributions All authors contributed significantly to the development and completion of this review study. The specific contributions of each author are as follows: Richard Mkechera [email protected] : writing—original draft; writing—review and editing. Sonika Dahiya ( [email protected] ): supervision Data Availability Statement : available upon request Funding : No fund Conflict of interest: The authors declare no competing interests. Ethical standard All authors have read, understood, and have complied as applicable with the statement on “Ethical responsibilities of Authors” as found in the Instructions for Authors. References Asudani, D. S., Nagwani, N. K., & Singh, P. (2024). Exploring the Performance of You Only Look Once Model for Object Detection in VQA Images. 2024 International Conference on Artificial Intelligence and Quantum Computation-Based Sensor Applications, ICAIQSA 2024 - Proceedings . https://doi.org/10.1109/ICAIQSA64000.2024.10882370 Bochkovskiy, A., Wang, C.-Y., & Liao, H.-Y. M. (2020). YOLOv4: Optimal Speed and Accuracy of Object Detection . http://arxiv.org/abs/2004.10934 Cortes, C., Vapnik, V., & Saitta, L. (1995). Support-Vector Networks Editor. In Machine Leaming (Vol. 20). Kluwer Academic Publishers. Dakari Aboyomi, D., & Daniel, C. (2023a). ITEJ Information Technology Engineering Journals A Comparative Analysis of Modern Object Detection Algorithms: YOLO vs. SSD vs. Faster R-CNN. ITEJ , 8 , 96–106. https://syekhnurjati.ac.id/journal/index.php/itej Dakari Aboyomi, D., & Daniel, C. (2023b). ITEJ Information Technology Engineering Journals A Comparative Analysis of Modern Object Detection Algorithms: YOLO vs. SSD vs. Faster R-CNN. ITEJ , 8 , 96–106. https://syekhnurjati.ac.id/journal/index.php/itej Emary, E., Elesawy, R. E., Abou El Ella, S. M., & Hassanien, A. E. (2014). Aquatic weeds prediction: A comparative study. Proceedings of 2014 9th IEEE International Conference on Computer Engineering and Systems, ICCES 2014 . https://doi.org/10.1109/ICCES.2014.7030969 Erftemeijer, P. L. A., & Robin Lewis, R. R. (2006). Environmental impacts of dredging on seagrasses: A review. Marine Pollution Bulletin , 52 (12), 1553–1572. https://doi.org/10.1016/j.marpolbul.2006.09.006 Everingham, M., Van Gool, L., Williams, C. K. I., Winn, J., & Zisserman, A. (2010). The Pascal visual object classes (VOC) challenge. International Journal of Computer Vision , 88 (2), 303–338. https://doi.org/10.1007/s11263-009-0275-4 Ge, Z., Liu, S., Wang, F., Li, Z., & Sun, J. (2021). YOLOX: Exceeding YOLO Series in 2021 . Girshick, R. (2015). Fast R-CNN . http://arxiv.org/abs/1504.08083 Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2013a). Rich feature hierarchies for accurate object detection and semantic segmentation . http://arxiv.org/abs/1311.2524 Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2013b). Rich feature hierarchies for accurate object detection and semantic segmentation . http://arxiv.org/abs/1311.2524 Gonzalez, R. C. ., & Woods, R. E. . (2018). Digital image processing . Pearson. Hao, Z., Lin, L., Post, C. J., & Mikhailova, E. A. (2024). Monitoring the spatial-temporal distribution of invasive plants in urban water using deep learning and remote sensing technology. Ecological Indicators , 162 . https://doi.org/10.1016/j.ecolind.2024.112061 Islam, M. S., & Islam, S. (n.d.). A COMPREHENSIVE ANALYSIS AND VARIOUS APPLICATIONS OF VEHICLE CLASSIFICATION USING DEEP LEARNING FROM THE PERSPECTIVE OF DHAKA CITY. In Scientific and Practical Cyber Security Journal (SPCSJ) (Vol. 9, Issue 1). https://www.researchgate.net/publication/391014664 Kabir, H., Juthi, T., Islam, M. T., Rahman, M. W., & Khan, R. (2024). WaterHyacinth: A comprehensive image dataset of various Water hyacinth species from different regions of Bangladesh. Data in Brief , 52 . https://doi.org/10.1016/j.dib.2023.109872 kaggle. (n.d.). WaterHyacinth: A Dataset of Water Hyacinth Species . Khanam, R., & Hussain, M. (2024). YOLOv11: An Overview of the Key Architectural Enhancements . http://arxiv.org/abs/2410.17725 Li, C., Li, L., Jiang, H., Weng, K., Geng, Y., Li, L., Ke, Z., Li, Q., Cheng, M., Nie, W., Li, Y., Zhang, B., Liang, Y., Zhou, L., Xu, X., Chu, X., Wei, X., & Wei, X. (2022). YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications . Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., & Berg, A. C. (2015). SSD: Single Shot MultiBox Detector . https://doi.org/10.1007/978-3-319-46448-0_2 Manojkumar, P. C., Kumar, L. S., & Jayanthi, B. (2023). Performance Comparison of Real-Time Object Detection Techniques with YOLOv4. Proceedings of 2023 International Conference on Signal Processing, Computation, Electronics, Power and Telecommunication, IConSCEPT 2023 . https://doi.org/10.1109/IConSCEPT57958.2023.10169970 Merlin, M. D. (2011). Seagrasses. Encyclopedia of Earth Sciences Series , Part 2 , 973–978. https://doi.org/10.1007/978-90-481-2639-2_146 Nikita Singhal, & Lalji Prasad. (2023). Comparative Analysis of Deep Learning-based Vehicle Detection Approaches. International Journal of Next-Generation Computing . https://doi.org/10.47164/ijngc.v14i2.976 Onyari, E. K., Fayomi, G. U., & Jaiyeola, A. T. (2024). Unveiling the situation of water hyacinth on freshwater bodies in Nigeria and South Africa: Management, workable practices and potentials. Case Studies in Chemical and Environmental Engineering , 10 . https://doi.org/10.1016/j.cscee.2024.100974 Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2015). You Only Look Once: Unified, Real-Time Object Detection . http://arxiv.org/abs/1506.02640 Redmon, J., & Farhadi, A. (2016). YOLO9000: Better, Faster, Stronger . http://arxiv.org/abs/1612.08242 Redmon, J., & Farhadi, A. (2018). YOLOv3: An Incremental Improvement . http://arxiv.org/abs/1804.02767 Ren, S., He, K., Girshick, R., & Sun, J. (n.d.). Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks . http://image-net.org/challenges/LSVRC/2015/results Shorten, C., & Khoshgoftaar, T. M. (2019). A survey on Image Data Augmentation for Deep Learning. Journal of Big Data , 6 (1). https://doi.org/10.1186/s40537-019-0197-0 Ultralytics. (n.d.-a). Comprehensive Guide to Ultralytics YOLOv5 . Ultralytics. (n.d.-b). YOLO8 . Wang, A., Chen, H., Liu, L., Chen, K., Lin, Z., Han, J., & Ding, G. (2024). YOLOv10: Real-Time End-to-End Object Detection . http://arxiv.org/abs/2405.14458 Wang, C.-Y., Bochkovskiy, A., & Liao, H.-Y. M. (2023). YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 7464–7475. https://doi.org/10.1109/CVPR52729.2023.00721 Wang, C.-Y., Yeh, I.-H., & Liao, H.-Y. M. (2024). YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information . http://arxiv.org/abs/2402.13616 Water Hyacinth (Eichhornia crassipes) Biology and Its Impacts on Ecosystem, Biodiversity, Economy and Human Well-being. (2019). Journal of Natural Sciences Research . https://doi.org/10.7176/jnsr/9-12-04 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-7323485\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":500673355,\"identity\":\"ce5b1813-ceac-47a8-9ce9-90f707b80e51\",\"order_by\":0,\"name\":\"Richard Mkechera\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Delhi Technological University (DTU)\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Richard\",\"middleName\":\"\",\"lastName\":\"Mkechera\",\"suffix\":\"\"},{\"id\":500673356,\"identity\":\"de41e6ca-5645-4d88-ab00-afea9de67211\",\"order_by\":1,\"name\":\"Sonika Dahiya\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/0lEQVRIiWNgGAWjYJCCgw0MDIwNEoxAqoIZKmZAtJYzzBJEaWGEaAGx2mBa8AD59t6DB2fUMMj2z25uk/g5z7rO4ADzww8MBXdwajE4cy7h4IZjDMYz7hxsk+zdli5hcIDNWILB4BluLRI5BgcfsDEkNtxIbJPg3XYYqIXBDCh+GLfDZoC0/GNInA/UIvl3DkgL+ze8WhhuALVsbGNI3ADUIs3bANLCg98WgzNnDA7O7JMw3ngjsdla5li65MzDPMUSCfgc1t5j/LHnm43svBvpD2++qbHm5zvevvHDhz94HAYB4NhggcQJKAEkENIABcwfiFQ4CkbBKBgFIwwAABy1W24cTkztAAAAAElFTkSuQmCC\",\"orcid\":\"\",\"institution\":\"Delhi Technological University (DTU)\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Sonika\",\"middleName\":\"\",\"lastName\":\"Dahiya\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-08-08 04:53:23\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-7323485/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-7323485/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":90881399,\"identity\":\"b32bd741-a499-47b2-8298-b5ca8367b54b\",\"added_by\":\"auto\",\"created_at\":\"2025-09-09 09:42:06\",\"extension\":\"jpg\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":83423,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eProposed pipeline\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"1.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/91cab8a8905f7761267ad71b.jpg\"},{\"id\":90881401,\"identity\":\"b2d50412-7c09-4611-812d-c8361f2b943b\",\"added_by\":\"auto\",\"created_at\":\"2025-09-09 09:42:06\",\"extension\":\"jpg\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":137710,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eThe training accuracy for FASTER R-CNN for WaterHyacinth datasets and the Kaggle Water Hyacinth dataset\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"2.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/efa03374499955b91f612d5b.jpg\"},{\"id\":90881402,\"identity\":\"abe17413-10e1-4124-b9e1-fb2264d72a43\",\"added_by\":\"auto\",\"created_at\":\"2025-09-09 09:42:06\",\"extension\":\"jpg\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":83723,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eTotal loss results for FASTER R-CNN for WaterHyacinth datasets and Kaggle Water Hyacinth dataset\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"3.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/bcb49ded6cfb054dacb571d6.jpg\"},{\"id\":90881405,\"identity\":\"3740982f-11f4-4e03-a817-b701950c18be\",\"added_by\":\"auto\",\"created_at\":\"2025-09-09 09:42:06\",\"extension\":\"jpg\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":77462,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eThe training accuracy for YOLOv11 for WaterHyacinth datasets and Kaggle Water Hyacinth dataset\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"4.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/2f09ae67b248dbd76f14c941.jpg\"},{\"id\":90881880,\"identity\":\"015c7d3a-6a05-4d49-93ba-f2acad74b954\",\"added_by\":\"auto\",\"created_at\":\"2025-09-09 09:50:06\",\"extension\":\"jpg\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":92446,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eThe training loss for YOLOv11 for both WaterHyacinth datasets and the Kaggle Water Hyacinth dataset\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"5.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/8c484a4b890b7e4b5d955b82.jpg\"},{\"id\":90881879,\"identity\":\"b255759b-aabe-4e3f-8e99-99b51c3de48a\",\"added_by\":\"auto\",\"created_at\":\"2025-09-09 09:50:06\",\"extension\":\"jpg\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":83433,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eModel accuracy comparison across datasets\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"6.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/94a84e546bd6172730def21b.jpg\"},{\"id\":90881404,\"identity\":\"a9206b40-d8c6-433d-8b28-988c2663c54a\",\"added_by\":\"auto\",\"created_at\":\"2025-09-09 09:42:06\",\"extension\":\"jpg\",\"order_by\":7,\"title\":\"Figure 7\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":206944,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eDetected Water Hyacinth\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"7.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/f49908f5416ad177264d8b9d.jpg\"},{\"id\":92663562,\"identity\":\"57c9f28d-9431-46f0-bbe4-03ca14413f6a\",\"added_by\":\"auto\",\"created_at\":\"2025-10-02 15:39:27\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1552400,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7323485/v1/e5f83c3d-2aa7-4c3c-a764-d3a577cb4fa8.pdf\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Towards Smart Monitoring of Invasive Aquatic Plants: A Comparative Evaluation of YOLOv11 and Faster R-CNN for Water Hyacinth Detection\",\"fulltext\":[{\"header\":\"1. Introduction\",\"content\":\"\\u003cp\\u003eWater resource conservation is a critical component of environmental sustainability, essential for preserving the biodiversity and ecological balance of aquatic ecosystems such as wetlands, lakes, and rivers. It includes the protection of both water quality and quantity, ensuring the long-term viability of freshwater systems. This aligns directly with the United Nations (UN) Sustainable Development Goal 6 (SDG 6), which emphasizes the need to ensure the availability and sustainable management of water and sanitation for all. Achieving this goal requires integrated efforts to monitor, manage, and safeguard freshwater ecosystems from degradation and invasive threats. These water bodies serve as a vital resource for drinking water, fisheries, agriculture, transportation, and habitat for aquatic organisms. Moreover, increasing anthropogenic pressure, like pollution, industrial discharge, invasive species like water hyacinth, and agricultural runoff have seriously degraded these ecosystems, threatening their ecological balance and the services they provide (Erftemeijer \\u0026amp; Robin Lewis, \\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e2006\\u003c/span\\u003e). Conservation efforts involve a combination monitoring of scientific monitoring, policy-making, community engagement, and application of modern technologies such as deep learning, remote sensing, and machine learning for effective management. By adopting sustainable practices and proactive restoration measures, water body conservation ensures the long-term availability and health of aquatic organisms, which are essential for human well-being and environmental stability(Merlin, \\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e2011\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003eAquatic weeds, generally invasive species such as water hyacinth \\u003cem\\u003e(Eichhornia crassipes)\\u003c/em\\u003e have caused significant challenges in freshwater ecosystems, and ecological and economic challenges (Emary et al., \\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e2014\\u003c/span\\u003e). Fast-growing plants can rapidly cover the surface of lakes, reservoirs, and rivers, block sunlight, and reduce oxygen levels in the water (\\u0026ldquo;Water Hyacinth (Eichhornia Crassipes) Biology and Its Impacts on Ecosystem, Biodiversity, Economy and Human Well-Being,\\u0026rdquo; 2019). Their dense mats hinder water flow, affect fishing activities, and clog irrigation canals together with hydroelectric systems, leading to an increase in maintenance costs and poor productivity (Onyari et al., \\u003cspan citationid=\\\"CR24\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e). Also, aquatic weeds can become breeding grounds for diseases like malaria, which leads to the contribution of public health risks.\\u003c/p\\u003e\\u003cp\\u003eHowever, proliferation normally starts from nutrient-rich water bodies by untreated water and agricultural runoff. Controlling the spread of water hyacinth requires integrated management strategies, and advanced monitoring and detection techniques, such as the application of deep learning and machine learning techniques which can enhance the elimination of these weeds. Object detection is the technique that locates and identifies objects within images or a video. Objects are classified based on their features such as shape, size, and color within a defined box known as a bounding box (Manojkumar et al., \\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e2023\\u003c/span\\u003e). However, in water hyacinth, object detection is applicable to automatically find and mark the presence of these invasive weeds in water bodies using ground-level or aerial images. By training deep learning models on labeled datasets of water hyacinth. These models can accurately detect and locate water hyacinths in both simple and complex environments. In recent years deep learning models have emerged as some of the most effective algorithms for object detection, significantly enhancing the accuracy of computer vision-based systems. To improve the detection of water hyacinths, this study explores the role of deep learning models in advancing the identification and monitoring of aquatic weeds (Hao et al., \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e). This work makes the following key contributions to the field of environmental monitoring:\\u003c/p\\u003e\\u003cp\\u003e\\u003col\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003eThis research work offers valuable insights for selecting suitable models for water hyacinth detection.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003eThis research work contributes a meticulously annotated dataset formatted for both COCO and YOLO standards, facilitating future studies in water hyacinth detection.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003cspan\\u003e\\u003cli\\u003e\\u003cp\\u003eThis research uses a dual-dataset approach to enhance the diversity and reliability of the evaluation framework.\\u003c/p\\u003e\\u003c/li\\u003e\\u003c/span\\u003e\\u003c/ol\\u003e\\u003c/p\\u003e\"},{\"header\":\"2. Deep Learning Models for Detecting Water Hyacinth\",\"content\":\"\\u003cp\\u003eTo support the automated detection of water hyacinths in remote sensing imagery, this study employs two state-of-the-art object detection models: YOLOv11 (Khanam \\u0026amp; Hussain, \\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e) and Faster R-CNN (Ren et al., n.d.). These architectures offer complementary strengths in speed and precision, respectively, which are critical for balancing real-time monitoring with detection accuracy in aquatic ecosystems. Object detection techniques have spearheaded the challenges of identification and localization of an image in real-world scenarios. The evolution of object detection solved a lot of the computer vision problems, some of the object detection which brought hope in the field of computer vision are Single Short Multibox Detector (SSD) (Liu et al., \\u003cspan citationid=\\\"CR20\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e), Region Proposal-Convolution Neural Network (R-CNN)(Girshick et al., \\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e), YOLO (You Only Look Once) (Redmon et al., \\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e). This paper chooses to work with YOLOv11 and faster R-CNN, to contribute to the strategic plans for the removal of water hyacinth.\\u003c/p\\u003e\\u003cp\\u003eAdvancement of the detection techniques leads to the solution of various problems existing in the community due to the high detection applicability on a particular object, although the detection of these models is highly affected by low resolution, occlusion, and complex environmental conditions (Islam \\u0026amp; Islam, n.d.). Deep learning models have been grouped into two major groups (1) Two-stage detector which at first identifies potential object location, this is done before object classification and bounding box regression, this architecture allows to detection of an object preciously but it's more resource intensive making it more complex for real-time application (2) a one stage detector processes the entire image in a single pass, allowing improved inference speed and lower memory usage while maintaining a competitive accuracy (Nikita Singhal \\u0026amp; Lalji Prasad, 2023). Understanding the performance and applicability of deep learning models toward the implementation of the embedded systems needs a detailed comparative analysis of different deep learning models in various environmental scenarios (Dakari Aboyomi \\u0026amp; Daniel, \\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e2023a\\u003c/span\\u003e), This makes it highly suitable for applications requiring real-time detection. This work highlights that model selection should comply with the specific need of the task, whether it is rapid real-time detection or high precision, this demonstrates that no single model is universally superior but rather depends on the performance of optimal requirements.\\u003c/p\\u003e\\u003cdiv id=\\\"Sec3\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e2.1 Two-stage detector\\u003c/h2\\u003e\\u003cp\\u003eThe two-stage detection architecture involves a sequential stage where the model generates a series of region proposals that represent areas of the image that are likely to present the object to be detected, classifies each region, and refines the corresponding bounding box (Asudani et al., \\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e). R-CNN (Girshick et al., \\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e) introduced a two-stage detector framework by applying selective search to propose regions, which were passed through a Convolutional Neural Network (CNN) to extract features of the object to be detected. Features were extracted using a Support Vector Machine (SVM) (Cortes et al., \\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e1995\\u003c/span\\u003e), and bounding box regressors were used to refine the locations. This model recorded a high accuracy compared to previous traditional models, but it was computationally expensive due to high computations on the overlapping regions. Since the introduction of R-CNN, there have been variants of R-CNN which were used to address the challenges faced by the R-CNN.\\u003c/p\\u003e\\u003cp\\u003eFast R-CNN (Girshick, \\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e), was proposed to address the challenges of R-CNN by integrating region proposal and feature extraction in a single Convolution Neural Network (CNN) forward pass, in this case, the entire image is processed by CNN only once; while generating a feature map and regions of interests are extracted from this region using pooling layer. This improvement of R-CNN reduced computation load and improved training speed to a large extent. Upon improving the efficiency of the Fast R-CNN, a new variant of R-CNN was introduced called Faster R-CNN (Ren et al., n.d.), this was more optimized when the Regional Proposal Network (RPN) was introduced, which shares a convolution feature with detection network and predicts region proposal directs by omitting the need for external proposal methods such as selective search. Faster R-CNN focuses more effectively on the complex scenario especially when the object varies in size and shape. Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e below shows architectural differences and improvements of R-CNN.\\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\\u003eTwo-Stage Detector: R-CNN\\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\\u003eRef\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eModel\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eReligion Proposal\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eDetection Network\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Girshick et al., \\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e2013\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eR-CNN\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eSelective search\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eSVM\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Girshick, \\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eFAST RCNN\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eSelective search\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eSoftMax\\u0026thinsp;+\\u0026thinsp;Bounding box regressor\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Ren et al., n.d.)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eFASTER RCNN\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eRegion Proposal Network (RPN)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eSoftMax\\u0026thinsp;+\\u0026thinsp;Bounding box regressor\\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=\\\"Sec4\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e2.2 One-stage detector\\u003c/h2\\u003e\\u003cp\\u003eWith the improvement of the two-stage object detector from R-CNN to faster R-CNN, object detection accuracy remained to be a very huge challenge. To solve all these challenges a stage detector was introduced. YOLO (You Only Look Once) (Redmon et al., \\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e), revolutionized real-time object detection by considering direct regression of the bounding boxes and class probabilities in a single pass through a neural network. YOLO reduces computational overhead while maintaining competitive accuracy, making them suitable for real-time applications. The invention of the YOLO algorithm marked the beginning of the journey of other YOLO versions, from YOLOv2 to YOLOv11, as shown in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e. YOLOv1 (Redmon et al., \\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e), introduced the journey of a single-stage object detector that defined it as a stage regression problem while maintaining high inference speed but failing to detect objects of small size and overlapping objects. The limitation of YOLOv1 was improved in YOLOv2 (YOLO9000)(Redmon \\u0026amp; Farhadi, \\u003cspan citationid=\\\"CR26\\\" class=\\\"CitationRef\\\"\\u003e2016\\u003c/span\\u003e), in this version they fixed the problem of detecting objects of small size by introducing anchor boxes and batch normalization.\\u003c/p\\u003e\\u003cp\\u003eYOLOv3(Redmon \\u0026amp; Farhadi, \\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e) greatly improved accuracy compared to YOLOv1 and YOLOv2, by introducing a backbone network which is Darknet-53 and multi-scale detection. To enhance feature extraction YOLOv4 (Bochkovskiy et al., \\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2020\\u003c/span\\u003e), was introduced with the addition of features like CSPDarknet53, PANeT for feature aggregation, and Mosaic augmentation which allows the model to see the object into multiple contexts and object scales within a single forward pass. To improve the performance of YOLOv4, YOLOv5(Jocher G et al, \\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e2020\\u003c/span\\u003e) was introduced and was greatly improved in terms of performance and accuracy where auto-learning bounding box anchors were introduced to improve accuracy and reduce the necessity of manual tunning, also inference speed was increased and training time was reduced compared to YOLOv4. Modularity and scalability were the key improvements of YOLOv5 where a number of multiple model sizes such as YOLOv5s, YOLOv5m, and YOLOv5l were introduced. YOLOX (Ge et al., \\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e), This version improved model generalization and simplified the model by introducing anchor-free design which makes it different from traditional anchor-based detection used in different versions of YOLO. Additionally, YOLOX applied a decoupled head for regression and classification, addressing the challenges observed in various versions of YOLO. Therefore, the improvement made in this model improved the accuracy of the YOLO model to a very great extent.\\u003c/p\\u003e\\u003cp\\u003eIndustrial applications and model deployment were very big questions due to the model size and performance in a real-time scenario, to improve model deployment capabilities and performance of a model YOLOv6(Li et al., \\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e2022\\u003c/span\\u003e) was introduced at which featured an anchor-free design, decoupled head architecture and improved training strategy. This model outperformed the previous version of YOLO due to high accuracy and precision. YOLOv7 (C.-Y. Wang et al., \\u003cspan citationid=\\\"CR32\\\" class=\\\"CitationRef\\\"\\u003e2023\\u003c/span\\u003e), introduced a re-parameterized convolution layer and extended efficiency layer aggregation efficiency network (E-ELAN). This model set up a new record for model accuracy and speed, therefore it became more useful for real-time object detection. YOLOv8(Ultralytics, \\u003cspan citationid=\\\"CR30\\\" class=\\\"CitationRef\\\"\\u003e2023\\u003c/span\\u003e), this version continues with the improvement of backbone redesigning, decoupled head, and segmentation tasks. This model is too flexible which makes it ideal for practical deployment across multiple domains. In YOLOv9 (C.-Y. Wang et al., \\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e), Generalized Efficient Layer Aggregation Network (GELAN) and programmable gradient information (PGI) were introduced. These significantly improved feature reuse and inference speed, especially for edge device problems. All versions of YOLO had a limitation on computation cost and inference speed, but in YOLOv10 (A. Wang et al., \\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e), they introduced a dual assignment for non-maximum suppression (NMS)-free training which resulted in low inference latency. To keep working on model size for easy deployment, YOLOv11(Khanam \\u0026amp; Hussain, \\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e) was introduced, where a unified neural network architecture that handles both bounding box regression and classification task was introduced, and in backbone network a newly cross-stage partial with spatial attention which improved accuracy due to the effectiveness of feature maps.\\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\\u003eOne-Stage Detector: YOLO\\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\\u003eRef\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eModel\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eImprovement\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eDetection Network\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Redmon et al., \\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv1\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eSingle Stage for classification and Regression problem\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eCNN (YOLO)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Redmon \\u0026amp; Farhadi, \\u003cspan citationid=\\\"CR26\\\" class=\\\"CitationRef\\\"\\u003e2016\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv2 (YOLO9000)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eAnchor boxes, Batch normalization, and High resolution\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eDarknet-19\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Redmon \\u0026amp; Farhadi, \\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv3\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eMulti-Scale prediction and residue blocks\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eDarknet-53\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Bochkovskiy et al., \\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2020\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eBag of freebies\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eCSPDarknet-53\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Jocher G et al, \\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e2020\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv5\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eModel scaling and Auto-learning bounding box\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eNew CSPDarknet-53\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Ge et al., \\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e2021\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOOX\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eAncho- free and decoupled head\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eModified CSPNet\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Li et al., \\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e2022\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv6\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eAnchor-free support and\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eEfficientRep\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(C.-Y. Wang et al., \\u003cspan citationid=\\\"CR32\\\" class=\\\"CitationRef\\\"\\u003e2023\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv7\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eELAN and extended reparameterization\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eELAN backbone\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Ultralytics, \\u003cspan citationid=\\\"CR30\\\" class=\\\"CitationRef\\\"\\u003e2023\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv8\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eAdvanced argumentation and new decoupled head\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eModified CSPnet\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(C.-Y. Wang et al., \\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv9\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eProgrammable gradient Information which improved accuracy\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eGELAN\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(A. Wang et al., \\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv10\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eLightweight design and enhanced real-time performance\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eOptimized CSPNet\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e(Khanam \\u0026amp; Hussain, \\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eYOLOv11\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eSpeed and Memory\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003eStage Partial with Spatial Attention (C2PSA)\\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\"},{\"header\":\"3. Proposed method\",\"content\":\"\\u003cp\\u003eThis study adopts an experimental design (Dakari Aboyomi \\u0026amp; Daniel, \\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e2023b\\u003c/span\\u003e) to examine the performance of FASTER R-CNN(Ren et al., n.d.) and YOLOv11(Khanam \\u0026amp; Hussain, \\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e) for water hyacinth detection. Two annotated datasets; the WaterHyacinth datasets (Kabir et al., \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e) and the Kaggle Water Hyacinth dataset (kaggle, n.d.). All datasets are preprocessed through proposed stages and trained through the training pipeline of each model with performance assessed across accuracy, speed, and computational efficiency metrics.\\u003c/p\\u003e\\u003cp\\u003eThe methodology emphasizes reproducibility, with detailed information on filtering techniques, augmentation strategies, and model training processes. The methodology is well illustrated in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e. The proposed pipeline aims to compare the performance of two object detection models; Faster R-CNN and YOLOv11 for the task of Water Hyacinth detection, the pipeline consists of multiple structured phases.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cdiv id=\\\"Sec6\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e3.1 Dataset acquisition\\u003c/h2\\u003e\\u003cp\\u003eThis work uses two distinct datasets; the WaterHyacinth dataset (Kabir et al., \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e) and the Kaggle Water Hyacinth dataset (kaggle, n.d.), which comprises varied resolution, lighting, and scene complexity for robust models' assurance as summarized on Table\\u0026nbsp;\\u003cspan refid=\\\"Tab3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e to indicate the total number of images in each dataset. These datasets are publicly available datasets, due to their quality and the number of images in each dataset, we had to add some preprocessing techniques in our working flow to improve the quality of images and the number of images in each dataset.\\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\\u003eDataset\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"3\\\"\\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\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colspan=\\\"3\\\" nameend=\\\"c3\\\" namest=\\\"c1\\\"\\u003e\\u003cp\\u003eDataset category\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eS.no\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eName of dataset\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eNumber of images\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e01\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eWaterHyacinth dataset (Kabir et al., \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e1000\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e02\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eKaggle Water Hyacinth datasets (kaggle, n.d.)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e584\\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=\\\"Sec7\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e3.2 Image Filtering\\u003c/h2\\u003e\\u003cp\\u003eThis study uses image filtering, with the aim of improving image quality by reducing noise. This involves loading zipped image files, extracting images from zipped files, and applying spatial filtering techniques where median filtering has been applied due to its excellence in noise reduction capability (Gonzalez \\u0026amp; Woods, \\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e2018\\u003c/span\\u003e). Finally, it stores all the filtered images for subsequent steps, as illustrated in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec8\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e3.3 Data argumentation\\u003c/h2\\u003e\\u003cp\\u003eAs shown in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e above, the number of images from both datasets is 584 for the Kaggle WaterHyacinth dataset and 1000 for WaterHyacinth. Dataset after doing a critical observation on both datasets, we realized that the size of data isn\\u0026rsquo;t enough to train a model. To enhance model generalization rotation argumentation technique was employed, which is done by exposing the images to the different orientations of the same objects. The rotation parameter was carefully selected; slight rotation was made at an angle of 20\\u0026deg;. A small angle is considered, because as the rotation angle increases data label is distorted, which may affect model generalization and accuracy of a model (Shorten \\u0026amp; Khoshgoftaar, \\u003cspan citationid=\\\"CR29\\\" class=\\\"CitationRef\\\"\\u003e2019\\u003c/span\\u003e). Following data augmentation, the total number of images increased to 2,000 and 1,732, respectively, as presented in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003e .\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab4\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 4\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eDataset after argumentation\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"3\\\"\\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\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colspan=\\\"3\\\" nameend=\\\"c3\\\" namest=\\\"c1\\\"\\u003e\\u003cp\\u003eDataset category\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eS.no\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eName of dataset\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eNumber of images\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e01\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eWaterHyacinth \\u003cb\\u003edataset\\u003c/b\\u003e (Kabir et al., \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e2000\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e02\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eKaggle Water Hyacinth datasets (kaggle, n.d.)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e1752\\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=\\\"Sec9\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e3.4 Data annotation and Splitting\\u003c/h2\\u003e\\u003cp\\u003eSince our study uses YOLOv11 and FASTER R-CNN. Therefore, there is a need to annotate images in two formats (1) YOLO format(Redmon et al., \\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e2015\\u003c/span\\u003e) and (2) COCO format(Everingham et al., \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2010\\u003c/span\\u003e). For YOLOv11, the annotation is well prepared using a specific YOLO format, which includes bounding box coordinates relative to the image dimension. For FASTER R-CNN, the annotation is structured in JSON COCO format enabling its compatibility and effectiveness in TensorFlow object detection API. The annotated data sets are then split into validation sets, Training sets, and testing sets in the ratio of 20:70:10 respectively, to ensure uniform label distribution across all sets (Everingham et al., \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e2010\\u003c/span\\u003e).\\u003c/p\\u003e\\u003c/div\\u003e\"},{\"header\":\"4. Implementation\",\"content\":\"\\u003cp\\u003eObject detection was implemented using YOLOv11 and FASTER R-CNN and the two different datasets: the Water Hyacinth dataset and the Kaggle Water Hyacinth dataset. Table\\u0026nbsp;\\u003cspan refid=\\\"Tab5\\\" class=\\\"InternalRef\\\"\\u003e5\\u003c/span\\u003e discusses the hardware and software requirements.\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab5\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 5\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eExperiment setup\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"3\\\"\\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\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e01\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eLanguage\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003ePython\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e02\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eEnvironment\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eGPU\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\" morerows=\\\"1\\\" rowspan=\\\"2\\\"\\u003e\\u003cp\\u003e03\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\" morerows=\\\"1\\\" rowspan=\\\"2\\\"\\u003e\\u003cp\\u003eDataset\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eWaterHyacinth dataset (Kabir et al., \\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e2024\\u003c/span\\u003e)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eKaggle Water Hyacinth datasets (kaggle, n.d.)\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e04\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003ePreprocessing\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eImage filtering\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e05\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eAugmentation\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eangle 20\\u0026deg;\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e06\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eArchitecture\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eYOLOv11 and FASTER R-CNN\\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\\u003eThis section details the experiment setup and step-by-step methodology employed to train and evaluate two distinct datasets using both YOLOv11 and FASTER R-CNN architectures. This comparative analysis follows a systematic approach to ensure reproducibility of the results at the end of the experiment. Each architecture was trained independently on both datasets as indicated on both \\u003cem\\u003eFig.\\u0026nbsp;2\\u003c/em\\u003e and \\u003cem\\u003eFig.\\u0026nbsp;4\\u003c/em\\u003e respectively, and performance metrics such as loss curve, training time, model size, and inference speed were recorded. In addition to training our model, we systematically monitored the training loss throughout the entire training process. This enabled us to assess the model's learning behavior over time and identify any signs of overfitting or underfitting. The recorded training loss trends are presented in Fig.\\u0026nbsp;3 and Fig.\\u0026nbsp;5 respectively, offering insights into the model\\u0026rsquo;s convergence and overall performance during training.\\u003c/p\\u003e\\u003cdiv id=\\\"Sec11\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e4.1 Analysis and comparison\\u003c/h2\\u003e\\u003cp\\u003eResult analysis is obtained by comparing the performance metrics of every architecture. Metrics considered during experimentation were inference speed, model size, and accuracy which gives a clear picture of which model is effective for water hyacinth detection. Since all datasets were split into test sets, then the test set was executed from both datasets, for the purpose of showing the effectiveness in detection and practical implication of each architecture. The analysis also aims to provide and suggest which model should be deployed to the embedded systems for easy and fast detection of water hyacinths in the aquatic environment.\\u003c/p\\u003e\\u003c/div\\u003e\\u003cdiv id=\\\"Sec12\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e4.2 Evaluation strategy\\u003c/h2\\u003e\\u003cp\\u003eThe study has considered the inference speed as one of the metrics to evaluate model performance. In deep learning models inference speed measures how faster, the model processes the input data and produces the output (Bochkovskiy et al., \\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2020\\u003c/span\\u003e). The formula for the inference speed is:\\u003cdiv id=\\\"Equ1\\\" class=\\\"Equation\\\"\\u003e\\u003cdiv format=\\\"TEX\\\" class=\\\"mathdisplay\\\" id=\\\"FileID_Equ1\\\" name=\\\"EquationSource\\\"\\u003e\\n$$\\\\:\\\\text{I}\\\\text{n}\\\\text{f}\\\\text{e}\\\\text{r}\\\\text{e}\\\\text{n}\\\\text{c}\\\\text{e}\\\\:\\\\text{s}\\\\text{p}\\\\text{e}\\\\text{e}\\\\text{d}=\\\\frac{\\\\text{T}\\\\text{o}\\\\text{t}\\\\text{a}\\\\text{l}\\\\:\\\\text{n}\\\\text{u}\\\\text{m}\\\\text{b}\\\\text{e}\\\\text{r}\\\\:\\\\text{o}\\\\text{f}\\\\:\\\\text{i}\\\\text{m}\\\\text{a}\\\\text{g}\\\\text{e}\\\\text{s}\\\\:}{\\\\text{T}\\\\text{o}\\\\text{t}\\\\text{a}\\\\text{l}\\\\:\\\\text{r}\\\\text{e}\\\\text{f}\\\\text{e}\\\\text{r}\\\\text{e}\\\\text{n}\\\\text{c}\\\\text{e}\\\\:\\\\text{t}\\\\text{i}\\\\text{m}\\\\text{e}\\\\:\\\\left(\\\\text{i}\\\\text{n}\\\\:\\\\text{s}\\\\text{e}\\\\text{c}\\\\text{o}\\\\text{n}\\\\text{d}\\\\:\\\\right)}$$\\u003c/div\\u003e\\u003cdiv class=\\\"EquationNumber\\\"\\u003e1\\u003c/div\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eIn addition to that, there are some metrics that have been considered in this study such as accuracy and precision. To ensure model prediction is robust, reliable, and good enough for the intended prediction, we must measure how accurate and precise the model is.\\u003c/p\\u003e\\u003cdiv id=\\\"Equ2\\\" class=\\\"Equation\\\"\\u003e\\u003cdiv format=\\\"TEX\\\" class=\\\"mathdisplay\\\" id=\\\"FileID_Equ2\\\" name=\\\"EquationSource\\\"\\u003e\\n$$\\\\:\\\\text{A}\\\\text{c}\\\\text{c}\\\\text{u}\\\\text{r}\\\\text{a}\\\\text{c}\\\\text{y}=\\\\frac{\\\\text{T}\\\\text{P}\\\\:+\\\\:\\\\text{T}\\\\text{N}\\\\:}{\\\\text{T}\\\\text{o}\\\\text{t}\\\\text{a}\\\\text{l}\\\\:\\\\text{s}\\\\text{a}\\\\text{m}\\\\text{p}\\\\text{l}\\\\text{e}\\\\text{s}}$$\\u003c/div\\u003e\\u003cdiv class=\\\"EquationNumber\\\"\\u003e2\\u003c/div\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eWhere TP is referred to as True Positive and TN is referred to as True Negative. From (2) accuracy is the ratio of correctly predicted samples or instances to the total number of predictions. This metric always gives a quick overview to the researcher on the behavioral balance of the classes; with this regard, we have used this metric in this study as a comparison baseline metric.\\u003cdiv id=\\\"Equ3\\\" class=\\\"Equation\\\"\\u003e\\u003cdiv format=\\\"TEX\\\" class=\\\"mathdisplay\\\" id=\\\"FileID_Equ3\\\" name=\\\"EquationSource\\\"\\u003e\\n$$\\\\:\\\\text{P}\\\\text{r}\\\\text{e}\\\\text{c}\\\\text{i}\\\\text{s}\\\\text{i}\\\\text{o}\\\\text{n}=\\\\frac{\\\\text{T}\\\\text{P}\\\\:}{\\\\text{T}\\\\text{P}\\\\:+\\\\:\\\\text{F}\\\\text{P}}$$\\u003c/div\\u003e\\u003cdiv class=\\\"EquationNumber\\\"\\u003e3\\u003c/div\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eWhere precision is defined as the ratio of True Positive and the total number of True Positive and False Positive, it answers the question, of how many samples are positive during the training, out of all predicted samples. It is more useful when understanding the correctness of the model.\\u003c/p\\u003e\\u003c/div\\u003e\"},{\"header\":\"5. Results and discussion\",\"content\":\"\\u003cp\\u003eThis section presents and analyzes results obtained from the comparative evaluation of the Faster R-CNN and YOLOv11 models across multiple performance metrics, including accuracy, precision, mean Average Precision (mAP), and inference speed. The evaluation was conducted using two different datasets with varied characteristics to assess the models' effectiveness under different conditions.\\u003c/p\\u003e\\u003cdiv id=\\\"Sec14\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003e5.1 Result by Dataset\\u003c/h2\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab6\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 6\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eWaterHyacinth dataset\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"3\\\"\\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\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eMetrics\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eFaster R-CNN\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cem\\u003eYOLOv11\\u003c/em\\u003e\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAccuracy\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e0.85\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e0.963\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003emAP\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e0.72\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e0.977\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eModel size\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e294.43MB\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e16.21MB\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eTraining time\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e60min\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e24.18min\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eInference speed\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e4.8ms\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e4.5ms\\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\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab7\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 7\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eKaggle Water Hyacinth dataset\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"3\\\"\\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\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eMetrics\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eFaster R-CNN\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eYOLOv11\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAccuracy\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e0.77\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e0.938\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003emAP\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e0.67\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e0.956\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eModel size\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e294MB\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e12.17MB\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eTraining time\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e55min\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e17.52min\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eInference speed\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e4.8ms\\u003c/b\\u003e\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e\\u003cb\\u003e4.4ms\\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\\u003cp\\u003eThe above results were obtained after training distinct datasets which were trained under the same settings. In this study, we have used multiple indicators (Accuracy, Mean Average Precision, model size, Training time, and inference speed) as the core metrics to evaluate Faster R-CNN and YOLOV11. The purpose of using the two different datasets is to increase the generalizability and clarity of findings on which model is useful for Water Hyacinth detection.\\u003c/p\\u003e\\u003cp\\u003eAs is seen in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003e and Table\\u0026nbsp;\\u003cspan refid=\\\"Tab7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003e respectively YOLOV11 significantly outperforms the performance of faster R-CNN in accuracy and computational efficiency over the two distinct datasets used in this study. The results explain how YOLOV11 is suitable for real-time Water Hyacinth detection applications, due to its low inference speed and small model size. The results suggest YOLOV11 can be useful particularly in mobile deployments and in edge-computing where speed and model size are seriously considered.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003eFigure\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003e, illustrates the comparative analysis of model accuracy between YOLOv11 and Faster R-CNN across two datasets. On both two sets of data YOLOv11, shows an outstanding performance in terms of accuracy. However, WaterHyacinth Dataset, YOLOv11 achieves an accuracy of 0.963, outperforming Faster R-CNN which achieves 0.85. For the Kaggle Water Hyacinth Datasets YOLOv11 attains 0.938 while Faster R-CNN scores a lower accuracy of 0.77. Figure\\u0026nbsp;5 shows that YOLOv11 has stronger generalization capacities and is more effective in detecting and classifying Water Hyacinth instances across different water bodies.\\u003c/p\\u003e\\u003cp\\u003eThe steep upward trend from Faster R-CNN to YOLOv11 in both lines reinforces YOLOv11's advantage in terms of accuracy. Thus, the graph provides very clear empirical evidence supporting YOLOv11, as the best and most accurate model for the detection of Water Hyacinth in various aquatic environments compared to Faster R-CNN.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe detection output shown in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003e, demonstrates the successful detection of water hyacinths across different visual scenarios, each input image contains a bounding box that accurately localizes Water Hyacinth clusters, regardless of their scale variation, background complexity, and lighting condition. All models demonstrate robustness to the environmental noises as they detect the target class reliably. This level of performance shows high spatial pressure and informs that the trained detector has effectively learned the distinguishing features of Water Hyacinth.\\u003c/p\\u003e\\u003c/div\\u003e\"},{\"header\":\"6. Conclusion and Future Work\",\"content\":\"\\u003cp\\u003eThis research work introduces a comparative analysis of YOLOv11 and FASTER R-CNN for Water Hyacinth detection using two distinct datasets. The evaluation centered on the key performance metrics such as accuracy, inference speed, training time, and mean average precision (mAP). The findings of this research shown in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003e and Table\\u0026nbsp;\\u003cspan refid=\\\"Tab7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003e respectively, demonstrate that YOLOv11 consistently outperforms Faster R-CNN in terms of training efficiency and inference speed, recording significantly faster detection times and reduced time taken to train a model. However, YOLOv11 exhibited superior and higher mAP across both Kaggle Water Hyacinth and Water Hyacinth datasets, highlighting its capability for real-time and resource-efficient deployment. In addition, while faster R-CNN delivers competitive accuracy, it scores higher computation cost, high memory size, and slower inference speed, becoming less suitable for embedded applications.\\u003c/p\\u003e\\u003cp\\u003eYOLOv11 emerges to be a more practical and scalable solution for real-time water hyacinth and other aquatic species detection and monitoring. The impact of the aquatic weeds in water sources is still very high and more challenging to many countries around the world, therefore this research work provides a third eye toward the process of eliminating aquatic weeds from water bodies. With these research findings, we suggest, to ensure the reliable elimination of water hyacinth in the future there should be an exploration of the integration of advanced data argumentation techniques such as Generative Adversarial Networks(GAN) and rotation-based transformation to increase the size of water hyacinth datasets which has become a very big challenge on implementing computer vision-oriented application to facilitate the elimination of the aquatic weeds, especially water hyacinth and to enhance model robustness under diverse environmental condition.\\u003c/p\\u003e\\u003cp\\u003eIn the future detection of water hyacinth may focus on extending the comparative evaluation of water hyacinth detection by incorporating lightweight transformer-based architectures, as well as exploring additional variants of the YOLOv11 model. Such models hold significant promise for enhancing performance, particularly in edge computing environments where computational resources are limited. The integration of real-time feedback mechanisms, linked with active learning frameworks, can further improve the model's adaptability to changing environmental conditions. This would facilitate continuous learning and ensure robust detection capabilities in dynamic and heterogeneous aquatic ecosystems. By adopting these advanced approaches, future systems could become more responsive, efficient, and scalable for widespread deployment in ecological monitoring and management applications.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eAuthor contributions\\u0026nbsp;\\u003c/strong\\u003eAll authors contributed significantly to the development and completion of this review study. The specific contributions of each author are as follows:\\u003c/p\\u003e\\n\\u003cul\\u003e\\n \\u003cli\\u003eRichard Mkechera \\u003cu\\u003erichardmmkechera_24phdse04@dtu.ac.in\\u003c/u\\u003e: writing\\u0026mdash;original draft; writing\\u0026mdash;review and editing.\\u0026nbsp;\\u003c/li\\u003e\\n \\u003cli\\u003eSonika Dahiya (sonika.dahiya@dtu.ac.in): supervision\\u0026nbsp;\\u003c/li\\u003e\\n\\u003c/ul\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e\\u0026nbsp;Data Availability\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003eStatement\\u003c/strong\\u003e\\u003cstrong\\u003e:\\u0026nbsp;\\u003c/strong\\u003eavailable upon request\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e\\u0026nbsp;Funding\\u003c/strong\\u003e: No fund\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e\\u0026nbsp;Conflict\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003eof\\u0026nbsp;\\u003c/strong\\u003e\\u003cstrong\\u003einterest:\\u0026nbsp;\\u003c/strong\\u003eThe\\u0026nbsp;authors\\u0026nbsp;declare\\u0026nbsp;no\\u0026nbsp;competing\\u0026nbsp;interests.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003e\\u0026nbsp;Ethical standard\\u0026nbsp;\\u003c/strong\\u003eAll authors have read, understood, and have complied as applicable with the statement on \\u0026ldquo;Ethical responsibilities of Authors\\u0026rdquo; as found in the Instructions for Authors.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\n\\u003cli\\u003eAsudani, D. S., Nagwani, N. K., \\u0026amp; Singh, P. (2024). Exploring the Performance of You Only Look Once Model for Object Detection in VQA Images. \\u003cem\\u003e2024 International Conference on Artificial Intelligence and Quantum Computation-Based Sensor Applications, ICAIQSA 2024 - Proceedings\\u003c/em\\u003e. https://doi.org/10.1109/ICAIQSA64000.2024.10882370\\u003c/li\\u003e\\n\\u003cli\\u003eBochkovskiy, A., Wang, C.-Y., \\u0026amp; Liao, H.-Y. M. (2020). \\u003cem\\u003eYOLOv4: Optimal Speed and Accuracy of Object Detection\\u003c/em\\u003e. http://arxiv.org/abs/2004.10934\\u003c/li\\u003e\\n\\u003cli\\u003eCortes, C., Vapnik, V., \\u0026amp; Saitta, L. (1995). Support-Vector Networks Editor. In \\u003cem\\u003eMachine Leaming\\u003c/em\\u003e (Vol. 20). Kluwer Academic Publishers.\\u003c/li\\u003e\\n\\u003cli\\u003eDakari Aboyomi, D., \\u0026amp; Daniel, C. (2023a). ITEJ Information Technology Engineering Journals A Comparative Analysis of Modern Object Detection Algorithms: YOLO vs. SSD vs. Faster R-CNN. \\u003cem\\u003eITEJ\\u003c/em\\u003e, \\u003cem\\u003e8\\u003c/em\\u003e, 96\\u0026ndash;106. https://syekhnurjati.ac.id/journal/index.php/itej\\u003c/li\\u003e\\n\\u003cli\\u003eDakari Aboyomi, D., \\u0026amp; Daniel, C. (2023b). ITEJ Information Technology Engineering Journals A Comparative Analysis of Modern Object Detection Algorithms: YOLO vs. SSD vs. Faster R-CNN. \\u003cem\\u003eITEJ\\u003c/em\\u003e, \\u003cem\\u003e8\\u003c/em\\u003e, 96\\u0026ndash;106. https://syekhnurjati.ac.id/journal/index.php/itej\\u003c/li\\u003e\\n\\u003cli\\u003eEmary, E., Elesawy, R. E., Abou El Ella, S. M., \\u0026amp; Hassanien, A. E. (2014). Aquatic weeds prediction: A comparative study. \\u003cem\\u003eProceedings of 2014 9th IEEE International Conference on Computer Engineering and Systems, ICCES 2014\\u003c/em\\u003e. https://doi.org/10.1109/ICCES.2014.7030969\\u003c/li\\u003e\\n\\u003cli\\u003eErftemeijer, P. L. A., \\u0026amp; Robin Lewis, R. R. (2006). Environmental impacts of dredging on seagrasses: A review. \\u003cem\\u003eMarine Pollution Bulletin\\u003c/em\\u003e, \\u003cem\\u003e52\\u003c/em\\u003e(12), 1553\\u0026ndash;1572. https://doi.org/10.1016/j.marpolbul.2006.09.006\\u003c/li\\u003e\\n\\u003cli\\u003eEveringham, M., Van Gool, L., Williams, C. K. I., Winn, J., \\u0026amp; Zisserman, A. (2010). The Pascal visual object classes (VOC) challenge. \\u003cem\\u003eInternational Journal of Computer Vision\\u003c/em\\u003e, \\u003cem\\u003e88\\u003c/em\\u003e(2), 303\\u0026ndash;338. https://doi.org/10.1007/s11263-009-0275-4\\u003c/li\\u003e\\n\\u003cli\\u003eGe, Z., Liu, S., Wang, F., Li, Z., \\u0026amp; Sun, J. (2021). \\u003cem\\u003eYOLOX: Exceeding YOLO Series in 2021\\u003c/em\\u003e.\\u003c/li\\u003e\\n\\u003cli\\u003eGirshick, R. (2015). \\u003cem\\u003eFast R-CNN\\u003c/em\\u003e. http://arxiv.org/abs/1504.08083\\u003c/li\\u003e\\n\\u003cli\\u003eGirshick, R., Donahue, J., Darrell, T., \\u0026amp; Malik, J. (2013a). \\u003cem\\u003eRich feature hierarchies for accurate object detection and semantic segmentation\\u003c/em\\u003e. http://arxiv.org/abs/1311.2524\\u003c/li\\u003e\\n\\u003cli\\u003eGirshick, R., Donahue, J., Darrell, T., \\u0026amp; Malik, J. (2013b). \\u003cem\\u003eRich feature hierarchies for accurate object detection and semantic segmentation\\u003c/em\\u003e. http://arxiv.org/abs/1311.2524\\u003c/li\\u003e\\n\\u003cli\\u003eGonzalez, R. C. ., \\u0026amp; Woods, R. E. . (2018). \\u003cem\\u003eDigital image processing\\u003c/em\\u003e. Pearson.\\u003c/li\\u003e\\n\\u003cli\\u003eHao, Z., Lin, L., Post, C. J., \\u0026amp; Mikhailova, E. A. (2024). Monitoring the spatial-temporal distribution of invasive plants in urban water using deep learning and remote sensing technology. \\u003cem\\u003eEcological Indicators\\u003c/em\\u003e, \\u003cem\\u003e162\\u003c/em\\u003e. https://doi.org/10.1016/j.ecolind.2024.112061\\u003c/li\\u003e\\n\\u003cli\\u003eIslam, M. S., \\u0026amp; Islam, S. (n.d.). A COMPREHENSIVE ANALYSIS AND VARIOUS APPLICATIONS OF VEHICLE CLASSIFICATION USING DEEP LEARNING FROM THE PERSPECTIVE OF DHAKA CITY. In \\u003cem\\u003eScientific and Practical Cyber Security Journal (SPCSJ)\\u003c/em\\u003e (Vol. 9, Issue 1). https://www.researchgate.net/publication/391014664\\u003c/li\\u003e\\n\\u003cli\\u003eKabir, H., Juthi, T., Islam, M. T., Rahman, M. W., \\u0026amp; Khan, R. (2024). WaterHyacinth: A comprehensive image dataset of various Water hyacinth species from different regions of Bangladesh. \\u003cem\\u003eData in Brief\\u003c/em\\u003e, \\u003cem\\u003e52\\u003c/em\\u003e. https://doi.org/10.1016/j.dib.2023.109872\\u003c/li\\u003e\\n\\u003cli\\u003ekaggle. (n.d.). \\u003cem\\u003eWaterHyacinth: A Dataset of Water Hyacinth Species\\u003c/em\\u003e.\\u003c/li\\u003e\\n\\u003cli\\u003eKhanam, R., \\u0026amp; Hussain, M. (2024). \\u003cem\\u003eYOLOv11: An Overview of the Key Architectural Enhancements\\u003c/em\\u003e. http://arxiv.org/abs/2410.17725\\u003c/li\\u003e\\n\\u003cli\\u003eLi, C., Li, L., Jiang, H., Weng, K., Geng, Y., Li, L., Ke, Z., Li, Q., Cheng, M., Nie, W., Li, Y., Zhang, B., Liang, Y., Zhou, L., Xu, X., Chu, X., Wei, X., \\u0026amp; Wei, X. (2022). \\u003cem\\u003eYOLOv6: A Single-Stage Object Detection Framework for Industrial Applications\\u003c/em\\u003e.\\u003c/li\\u003e\\n\\u003cli\\u003eLiu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., \\u0026amp; Berg, A. C. (2015). \\u003cem\\u003eSSD: Single Shot MultiBox Detector\\u003c/em\\u003e. https://doi.org/10.1007/978-3-319-46448-0_2\\u003c/li\\u003e\\n\\u003cli\\u003eManojkumar, P. C., Kumar, L. S., \\u0026amp; Jayanthi, B. (2023). Performance Comparison of Real-Time Object Detection Techniques with YOLOv4. \\u003cem\\u003eProceedings of 2023 International Conference on Signal Processing, Computation, Electronics, Power and Telecommunication, IConSCEPT 2023\\u003c/em\\u003e. https://doi.org/10.1109/IConSCEPT57958.2023.10169970\\u003c/li\\u003e\\n\\u003cli\\u003eMerlin, M. D. (2011). Seagrasses. \\u003cem\\u003eEncyclopedia of Earth Sciences Series\\u003c/em\\u003e, \\u003cem\\u003ePart 2\\u003c/em\\u003e, 973\\u0026ndash;978. https://doi.org/10.1007/978-90-481-2639-2_146\\u003c/li\\u003e\\n\\u003cli\\u003eNikita Singhal, \\u0026amp; Lalji Prasad. (2023). Comparative Analysis of Deep Learning-based Vehicle Detection Approaches. \\u003cem\\u003eInternational Journal of Next-Generation Computing\\u003c/em\\u003e. https://doi.org/10.47164/ijngc.v14i2.976\\u003c/li\\u003e\\n\\u003cli\\u003eOnyari, E. K., Fayomi, G. U., \\u0026amp; Jaiyeola, A. T. (2024). Unveiling the situation of water hyacinth on freshwater bodies in Nigeria and South Africa: Management, workable practices and potentials. \\u003cem\\u003eCase Studies in Chemical and Environmental Engineering\\u003c/em\\u003e, \\u003cem\\u003e10\\u003c/em\\u003e. https://doi.org/10.1016/j.cscee.2024.100974\\u003c/li\\u003e\\n\\u003cli\\u003eRedmon, J., Divvala, S., Girshick, R., \\u0026amp; Farhadi, A. (2015). \\u003cem\\u003eYou Only Look Once: Unified, Real-Time Object Detection\\u003c/em\\u003e. http://arxiv.org/abs/1506.02640\\u003c/li\\u003e\\n\\u003cli\\u003eRedmon, J., \\u0026amp; Farhadi, A. (2016). \\u003cem\\u003eYOLO9000: Better, Faster, Stronger\\u003c/em\\u003e. http://arxiv.org/abs/1612.08242\\u003c/li\\u003e\\n\\u003cli\\u003eRedmon, J., \\u0026amp; Farhadi, A. (2018). \\u003cem\\u003eYOLOv3: An Incremental Improvement\\u003c/em\\u003e. http://arxiv.org/abs/1804.02767\\u003c/li\\u003e\\n\\u003cli\\u003eRen, S., He, K., Girshick, R., \\u0026amp; Sun, J. (n.d.). \\u003cem\\u003eFaster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks\\u003c/em\\u003e. http://image-net.org/challenges/LSVRC/2015/results\\u003c/li\\u003e\\n\\u003cli\\u003eShorten, C., \\u0026amp; Khoshgoftaar, T. M. (2019). A survey on Image Data Augmentation for Deep Learning. \\u003cem\\u003eJournal of Big Data\\u003c/em\\u003e, \\u003cem\\u003e6\\u003c/em\\u003e(1). https://doi.org/10.1186/s40537-019-0197-0\\u003c/li\\u003e\\n\\u003cli\\u003eUltralytics. (n.d.-a). \\u003cem\\u003eComprehensive Guide to Ultralytics YOLOv5\\u003c/em\\u003e.\\u003c/li\\u003e\\n\\u003cli\\u003eUltralytics. (n.d.-b). \\u003cem\\u003eYOLO8\\u003c/em\\u003e.\\u003c/li\\u003e\\n\\u003cli\\u003eWang, A., Chen, H., Liu, L., Chen, K., Lin, Z., Han, J., \\u0026amp; Ding, G. (2024). \\u003cem\\u003eYOLOv10: Real-Time End-to-End Object Detection\\u003c/em\\u003e. http://arxiv.org/abs/2405.14458\\u003c/li\\u003e\\n\\u003cli\\u003eWang, C.-Y., Bochkovskiy, A., \\u0026amp; Liao, H.-Y. M. (2023). YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors. \\u003cem\\u003e2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)\\u003c/em\\u003e, 7464\\u0026ndash;7475. https://doi.org/10.1109/CVPR52729.2023.00721\\u003c/li\\u003e\\n\\u003cli\\u003eWang, C.-Y., Yeh, I.-H., \\u0026amp; Liao, H.-Y. M. (2024). \\u003cem\\u003eYOLOv9: Learning What You Want to Learn Using Programmable Gradient Information\\u003c/em\\u003e. http://arxiv.org/abs/2402.13616\\u003c/li\\u003e\\n\\u003cli\\u003eWater Hyacinth (Eichhornia crassipes) Biology and Its Impacts on Ecosystem, Biodiversity, Economy and Human Well-being. (2019). \\u003cem\\u003eJournal of Natural Sciences Research\\u003c/em\\u003e. https://doi.org/10.7176/jnsr/9-12-04\\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\":\"info@researchsquare.com\",\"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\":\"Water Hyacinth, YOLOv11, Faster R-CNN, Inference speed, detection, invasive aquatic plants\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-7323485/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-7323485/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eIn recent years, the proliferation of water hyacinths across various freshwater bodies has become an escalating environmental concern due to its adverse effects on agricultural productivity, aquatic ecosystems, and public health. As a highly invasive aquatic weed, water hyacinth rapidly spreads, obstructing irrigation canals, drainage systems, and fish farms, depleting oxygen levels, and degrading water quality, thereby directly challenging the objectives of Sustainable Development Goal (SDG) 6, which calls for improved water quality and the protection of water-related ecosystems. Early detection is essential for controlling its spread, yet accurately distinguishing water hyacinths from other aquatic vegetation remains technically difficult. To address this, this study evaluates the effectiveness of deep learning models in the rapid and accurate detection of water hyacinths. Specifically, we conduct a comparative analysis of YOLOv11n and Faster R-CNN using two datasets: a custom WaterHyacinth dataset and the Kaggle Water Hyacinth dataset. The models are assessed based on accuracy, inference speed, model size, and mean average precision (mAP). Our findings indicate that YOLOv11n outperforms Faster R-CNN, achieving 96.3% and 93.8% accuracy, on the respective datasets. Furthermore, YOLOv11n\\u0026rsquo;s lightweight architecture and faster inference speed make it more suitable for real-time applications, highlighting its potential as a practical tool for the early detection and management of water hyacinth, thus contributing to SDG 6\\u0026rsquo;s mission of ensuring clean water and healthy freshwater ecosystems.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Towards Smart Monitoring of Invasive Aquatic Plants: A Comparative Evaluation of YOLOv11 and Faster R-CNN for Water Hyacinth Detection\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-09-09 09:42:01\",\"doi\":\"10.21203/rs.3.rs-7323485/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"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\":\"37e7fd6f-61cd-4a94-9231-e9c7cda65526\",\"owner\":[],\"postedDate\":\"September 9th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2025-10-02T15:38:13+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2025-09-09 09:42:01\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-7323485\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-7323485\",\"identity\":\"rs-7323485\",\"version\":[\"v1\"]},\"buildId\":\"8U1c8b4HqxoKbykW_rLl7\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}