{"paper_id":"3d902be4-d53e-4e20-830a-6c15e7ff8267","body_text":"Improved CNN System for Face Mask Recognition | 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 Improved CNN System for Face Mask Recognition Ammar Hussein Jassim, Ahmed Altaie, Amal Sufiuh Ajrash This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4251321/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 Deep learning, especially convolutional neural networks, has significantly improved performance in computer vision. Therefore, we designed and developed a modified deep convolutional neural network framework for detecting mask in facial images in a sizable synthesized and un-synthesized face mask dataset. The suggested method can be utilized to detect face masks in any image with a low-resolution, different alignments, complex, and noisy background by tuning the hyperparameters to accurately identify the existence of masks without generating overfitting. The experimentally obtained results demonstrate that the suggested model exhibits a significant efficiency level, achieving 97.39% accuracy, 97.34% precision, 97.41% recall, 97.37% F1-score, and 97.4% AUC. The empirical results have been documented after 35 iterations using optimized hyperparameter settings, and those predictive models were trained on 64,398 images with a 98% accuracy rate and 0.05 loss, proving the proposed work's reliability and robustness. Face Mask Detection Face Mask Recognition Deep Learning (DL) Convolutional Neural Network (CNN) Machine Learning (ML) Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 1. Introduction Artificial intelligence (AI) is an everyday occurrence due to technological innovations. AI is being used in various applications, from financial markets [ 1 ] virtual assistants driven by AI to self-driving automobiles [ 2 ]. In some areas, like computer vision, the growth of artificial intelligence is extremely astonishing. It allows machines to observe items in a way humans do [ 3 ], allowing them to identify, evaluate, and categorize objects. This is made possible using CNN, a dependable methodology that generates novel prospects for computer vision applications [ 4 ]. An algorithm for deep learning can assess images and accurately recognize and classify objects in them [ 5 ]. CNN is built in a way similar to how the brain works, and it can do things like analyze pixel data. Researchers were significantly assisted by deep learning computational methods to extract important features that best depict the issue [ 6 ]. Indeed, in several domains, including image classification, neural network techniques have been successfully applied, which combine technologies such as speech, face recognition, self-driving cars, cancer detection [ 7 , 8 ], etc. Deep learning demonstrated its effectiveness in detecting various object types for all of these applications. For the task of detecting the presence of masks, a lot of techniques have been developed from scratch [ 9 , 10 ], like the proposed model that will be introduced in these papers. Consequently, face and mask detection systems based on artificial intelligence are growing in popularity now [ 11 , 12 , 13 ]. This study proposes developing a system capable of promptly detecting the presence or absence of masks on faces by employ the CNN net architecture to extract distinctive features from the images. The proposed work aims to make these contributions: Design an efficient face mask classifier that detects any face present within the image regardless of its alignment and improves the accuracy and response time consumption. Adapt the model with high performance on high and small-size masked faces with different resolutions, formats, and blurry images. A model that can utilize CPU cores and hyperparameters tuner advantages. A sizable dataset is required to complete this challenge and train deep-learning models to recognize whether people wear masks. 2. Related Work This section includes the most recent and pertinent academic research on face-mask identification using CNN models. Dozens of journal papers were obtained, and all the papers [ 14 ] to [ 25 ] were selected for related review because of their detailed explanations, excellent experiments, and good discussions. Some papers are not considered because they don't have any experiments, especially ones without quantitative results; visual detection results aren't shown; or the dataset doesn't have enough images, say 1000 or more. Certain writing works, in particular, use similar methods and only try algorithms on different data sets. As a result, only those with large datasets and good performance were chosen for this work. Wadii B. et al., 2021 [ 14 ] recommend using MobileNetV2 [ 15 ] for facemask recognition in real-time. Several experiments show the performance of the suggested model (test and training accuracy is 99%). The dataset contains 21407 images. For training, use 17125 images; for testing, use 4282 images. Nagrath et al., 2021 [ 16 ] proposed SSDMNV2 as a model that uses a single-shot multi-box (SSD) and a MobilenetV2 architecture for face detection, which serves as the classifier's framework. To create a balanced dataset, it uses a dataset with 5521 images with “with_mask” labels and 5521 images without “without_mask” labels. The suggested model has an F1 score of 9.3% and an average accuracy of 92.64%. Dey S. et al., 2021 [ 17 ] used MobileNetV2 as a classifier, which relies on how well ResNet-10 [ 18 ] and SSD find ROI. They used two distinct datasets. The first dataset had 3835 samples from two classes (1916 with mask and 1919 without mask), and dataset 2, consisting of 1376 images, was used to assess the model's detection capabilities from images and a video stream. Experiment results indicate that with 20% validation samples, MobileNetV2 achieves a 93% accuracy rate for dataset 1, whereas it achieves an accuracy close to 100% for dataset 2. Sravani N. et al., 2022 [ 19 ], Authors proposed the MobileNet Mask system, a multi-phase recognition technique based on deep learning. The results show that MobileNet achieves around 93% accuracy using 770 validation images and approximately 100% accuracy with 276 validation samples. Sagar A. et al., 2022 [ 20 ], Authors describe a system using OpenCV, Keras packages, and MobileNetV2. 70% of the dataset was used for training and 30% for testing. The obtained accuracy of the system is 96.07%. Bose S. et al., 2023 [ 21 ] The proposed system uses YOLOv3 architecture and computer vision. This system is evaluated using the COCO dataset [ 22 ] containing 5,000 face-masked and unmasked pictures. Testing data is 20%, and training data is 80%. Experimental results show that the proposed system predicts mask-wearers with 99.2% accuracy and an F1-score of 0.99. Naufal M. A. et al., 2023 [ 23 ] This study demonstrated that a CNN model with OpenCV, Keras, and TensorFlow can be trained to detect facial masks in low illumination conditions using a recreational low-light image, producing an accuracy of 97.89%. Divide the data into training and test sets with a distribution ratio 70:30. consist of 714 with_mask images and 708 without_mask images; they also include 24 low-illumination images of with_mask and 22 of without_mask. Ritesh T. et al., 2023 [ 24 ] proposed an MTCNN system based on deep learning. Within the 1539 samples that comprise the dataset, 10% are used for validation, while 90% are for training. The suggested dataset was divided into with and without mask images, and 308 images were used for testing the system. The suggested model classifies 172 as masked images and 136 as without mask images, with 99.76% training and 99.35% testing accuracy, respectively. Sathwika B. et al., 2023 [ 25 ] suggested ResNet-CNN, and basic machine learning tools like OpenCV, TensorFlow, Keras, and Scikit-Learn were used. There are 1376 images that CNN used to train its system, and it can make 98% accurate predictions. 3. Dataset Description & Preparation A sizable dataset is required for this purpose to train deep learning models to distinguish between people with masks and those who are not. In this work, we collect our datasets from multiple sources. Indeed, our datasets are made up of a mix of freely available datasets and data that we got from the Google website. The idea was to make it diverse and unbiased. The images are various sizes, the faces are real and sketches, and the pictures are taken at different angles. It includes the complex images of sophisticated face masks printed with printed patterns, textures, or brands. Also, the image has faces that are partially obscured, such as by a beard, hat, hair, hijab, sunglasses, and hands covering the face. Figure 1. shown some images of non-mask-wearing individuals while Fig. 2. shown people using face masks. The dataset contains two folders containing images in jpg format. There are 85,864 images belonging to 2 classes: 46.735 with masks and 39.129 without masks. For evaluation, 75% can be primarily used to train face mask classifiers and 25% for testing by randomly selecting images from the dataset. Figure 3. shows a distribution of the dataset. Likewise, Figure 4. presents the percentage of each class’s size. 4. Proposed Model Architecture The proposed work employs deep neural networks, which are more effective than alternative classification algorithms [ 6 ]. However, training a deep neural network is extremely expensive due to its computationally intensive and time-consuming nature. Therefore, the proposed model utilizes the Num worker (processes) in CPU cores for data loading and GPU capabilities. In a sense, the CPU loads data more efficiently if the Num workers are greater than 1. Therefore, the GPU must wait less, especially when the large dataset is put in different folders on different drives. Therefore, the proposed model used two workers when preprocessing and augmenting the images. When preparing a dataset, input images from this dataset may need to be preprocessed and augmented. For this purpose, the proposed model, instead of processing input images one by one, replicates the pre-processing and augmentation tasks into units that can be executed concurrently, batching them together for parallel processing (multi-threading). Pre-processes a group of data from the dataset. As a result of the fact that input images are not dependent on one another, each thread can handle a subset of the input images, enabling operations to be parallelized across multiple cores of the CPU. This can help exploit the parallelism offered by modern CPUs and optimize the augmentation and pre-processing pipeline. After preparing the dataset, design the proposed system, which includes two parts. The first one is to detect faces and specify the ROI of detected faces for extraction. The second part is a CNN, which considers the main part of the system structure to extract features of faces and convert them into a feature map for classification. The affine transformation is used to detect facial features because there may be changes in face size and orientation in the clipped ROI. The subsequent subsections include a comprehensive depiction of each activity within the proposed architecture: 4.1 Image Preprocessing and Augmentation Within the context of this research, a series of steps are involved in preparing the images for training tasks such as normalization, resizing, and augmentation techniques such as rescale, height shift, width shift, rotation, zoom range, and horizontal flip. 4.2 Haar Cascade classifier The input image may vary in size, have multiple faces in a single image, and have non-frontal faces. The initial phase of our approach is to identify individual faces from one image by implements a haar cascade classifier with pre-trained XML files in OpenCV Python to detect faces. Figure 5 . shows the first part of the proposed system. 4.3 Convolutional Layer Firstly, employ a CNN with an initial hyperparameter configuration and input images processed through a sequence of operations. The precise structure of the proposed networks is shown in Fig. 6 . The CNN output categorizes the faces as either having a mask or not. The suggested model has five convolutional layers, two layers are fully connected, and one dense Softmax classifier at the last layer. The 3x3 kernel sizes used in each convolutional layer were placed on top. The convolutional layers comprise 16, 32, 64, and 128, and the system utilizes 256 kernel filters, each having a size of 3x3. As the image goes through the convolutional layers, it changes from an RGB image with 3 depth levels to 96x96x3. The output obtained from each convolution layer is subsequently sent to the max pooling layer with a stride of 2, where pooling is performed using a (2x2) window. The last levels of convolution layers are completely linked hidden layers (flattened layer) of 2304 units, followed by fully connected dense layers containing 1024 units and a dense softmax output layer of 2 units. To prevent overfitting, the 0.2 dropout is used to drop layers at random. 4.3.1 Activation Function In the proposed CNN, Rectified Linear Unit (ReLU) activation is applied following the convolutional layers, and softmax is a decider in the last layer. 4.3.2 Pooling Layer Pooling works on the concept of down-sampling. It reduces the complexity and dimensionality of further layers and eliminates noisy activations. MaxPooling with 2x2 filters and stride 2 is the most common type, and it is the one that is applied in the proposed model. 4.3.3 Hyperparameter Optimization Our goal should be to find the best set of hyperparameter values to get the perfect prediction results from our model. Thus, the suggested model's performance is evaluated on a separate validation dataset to prevent overfitting and select the best hyperparameters depending on the score of the validation results in an iterative process. So, it's important to consider the specific characteristics of the used dataset and problem domain when optimizing the hyperparameters. The proposed model uses Python automated libraries (GridSearchCV in the Scikit API library) for systematic hyperparameter optimization. This method examines a range of hyperparameter combinations (learning rate, batch size, epochs, and dropout) one by one, passes them to the model, and evaluates their performance to find the optimal set that maximizes the model's performance. 4.3.4 CNN Model Visualization The modified model is shown in Fig. 7 . takes an input of shapes 96,96,3 from the previous stage and, at the output, classifies each face in the image as a mask or without a mask image. 5. Result And Analysis In this work, the proposed dataset contains 39129 images with mask faces and 46735 images without mask faces, divided into training and test sets. There is an imbalance of classes in the training data. For testing purposes, the dataset contains 21466 images, 11638 instances of \"mask,\" and 9828 instances of \"without_mask.\" Evaluating and predicting real-world images was proposed to ensure the modified model's effectiveness. Once the classification model has been built and fitted to the training data, it is necessary to evaluate the model's performance and specify how well it generalizes to the tested data by making predictions on the test data using the prediction method. Therefore, the focus is mainly on classification issues by reporting different metrics for each class in the dataset. For this purpose, a classification report offers valuable information about the model’s performance for each class. Table 1 Describe the performance of the model in detail. Table 1 Classification report. Class Precision Recall F1-score support Mask 0.9796 0.9720 0.9758 11638 Without_mask 0.9671 0.9761 0.9716 9828 accuracy 0.9739 21466 Macro avg 0.9734 0.9740 0.9737 21466 Weighted avg 0.9739 0.9739 0.9739 21466 Our model was trained on 35 epochs, and as seen from Table 2 , we got a training accuracy of 98.32% with a training loss of 5.06% and a testing accuracy of 97.39% with a test loss of 17.50%, meaning that 97% of the predictions were correct. The obtained accuracy is not relative to a particular class, but performance over all classes, and a low mean squared error was also calculated to be equal to 0.26, which proves the proposed classifier is closer to forecasting the true value, which is quite good considering the size and complexity of the dataset and the good design of the proposed system architecture. Table 2 Performance evaluation for each epoch. Epoch Loss Accuracy Val_loss Val_accuracy 0 0.563913 0.738605 0.290697 0.876087 1 0.289507 0.877378 0.215679 0.913276 . . . . . . . . . . . . . . . 30 0.051584 0.981471 0.121372 0.972904 31 0.058623 0.981617 0.139657 0.970730 32 0.050712 0.982374 0.131059 0.971739 33 0.55145 0.983135 0130244 0.973758 34 0.050596 0.983287 0.130150 0.974730 Furthermore, as Fig. 8 . shows, our training history plot is smooth, Confirming the absence of overfitting within the training phase. The tested proposed model has no overlap of results and can discriminate between images with and without a mask with precision, recall, and an F1-score of around 97% for both classes, indicating balanced performance across both classes. Table 1 , 2 shows the different scores that we calculated for the improved model. The accuracy of positive predictions for the “Mask” class scored 0.9796, indicating that ≈ 98% of the masked images predicted as “Mask” are correct. Regarding the class \"Without_mask,\" the precision is 0.9671, indicating that ≈ 97% of the unmasked images predicted as “Without_mask” are correct. Clearly, the modified classifier returns decent results with high true positives. In contrast, the recall is high, equal to around 97%, which indicates the model returns most positive instances, minimizing false negatives but making fewer false positive errors, with some false positives as accrued in the confusion matrix in Fig. 9 . The \"Mask\" class has an F1-score of 0.9758, balancing precision and recall. Regarding the \"Without_mask\" category, the F1-score is 0.9716. This value indicates that the enhanced model performs well at predicting whether or not faces will wear masks. Therefore, a higher F1 score indicates a better overall performance. The classification report accuracy, considering the class distribution equal to 0.9734, is a macro-average of the accuracies for each class rather than calculating overall accuracy, which can be more informative in the case of imbalanced classes. Figure 9 . shows the confusion matrix we constructed to measure performance measures. The model can accurately categorize 11,312 out of the 11,638 test pictures using a mask. This is also visible by the high sensitivity value of class \"mask\"—just 97%. Basically, the suggested model classifies most of the “mask” and ‘without_mask’ images correctly at an acceptable performance. Another diagnostic tool commonly used is the region under the arc and the precision-recall curve. As seen in Fig. 10 , the curve bows towards the true positive rate (upper left corner (1,1)), which means the model has a good performance for predicting the positive classes. Figure 10 shows the AUC that signifies the area below the ROC curve, which falls between 0 and 1. It measures the entire performance of the proposed model. The obtained AUC score indicates improved model performance, which has the lowest false positive rates (FPR) and the highest true positive rates (TPR). Also, shows the perfect test of the modified model, passing through the left upper corner. It gives the impression of precision, with a recall value of approximately 99%. The PR curve in Fig. 11 . shows the performance of the proposed model (purple) and allows for a quick assessment of good prediction performance. It gives an impression of the tradeoff between precision and recall. The contrast between the obtained precision and recall is very small. As seen in the upper right corner, the curve is getting close to optimal performance. The good test has a PR curve that bows to the right upper corner (where 100% precision and 100% recall are at the peak point (1,1)). These curves give us the shape we would expect for perfect performance. A high space under the curve shows both high recall and precision. The threshold at point (0, 0) is set to 1.0. This indicates that the model does not discriminate between images with and without masks. An ideal classifier with good precision is associated with a low percentage of FPR, and recall relates to a low percentage of FNR. For further comparisons, the kappa and MCC (Matthews correlation coefficient) are taken into consideration as evaluation metrics. The model achieved a 0.947395 kappa result, representing a good agreement between the classification of observed classes and expected classes in the proposed dataset. The obtained value of the MCC is 0.947429, which indicates the proposed model can predict a high percentage of true positives and true negatives. Finally, the performance values obtained more accurately represent the true ability of the classifier to distinguish between the two classes. 6. Output 6.1 Evaluation images 6.2 Testing images 6.3 Comparison with State-of-the-Art Detectors The comparison results among the proposed model and other recent DL face mask detectors are shown in Table 3 summarizes the experiment’s findings and indicates that the suggested model had significantly higher accuracy and outperformed previous techniques in terms of accuracy. All the models were trained and tested on different datasets, environments, and network parameters. Table 3 Comparative analysis of various models. Literatures Network Dataset Results Proposed method DCNN Collected 85,864 images 97.47 training 97.39 testing accuracy (Nagrath P. et al.,2021) [ 16 ] DNN, MobilenetV2 690 Masked 686 Not-Masked 92.64 (Dey S. et al., 2021) [ 17 ] SSD and ResNet-10, MobileNetV2 3835 real images (IDS1), 1376 simulated images (IDS2) IDS1 Accuracy = 93%, IDS2 Accuracy = 100% (Bingshu W. et al.,2021) [ 26 ] Faster R-CNN, BLS Wearing-Mask-Detection (WMD) Recall = 93.54%, Precision = 94.84%, F1 = 94.19% (Jonathan S. et al.,2021) [ 27 ] MobileNetV2 13359 images, self-built Accuracy = 99:65% (AL-Dmour H. et al.,2023) [ 28 ] CNN RMFD, CFR, Masked-FaceNet Accuracy:99.5 (Mohamed L. et al.,2021) [ 29 ] ResNet-50, YOLO5 Masked Face Dataset (SMFD, RMFD) 785 Masked 785 Not-Masked Accuracy:99.49 (Vaibhav J. et al.,2022) [ 30 ] CNN RMFD (Real World Masked Face Dataset) and Kaggle Dataset Accuracy:96 (Francesco M. et al.,2021) [ 31 ] MobilenetV2 2165 Masked 1930 Not Masked Accuracy:98.0 (Sethi, S. et al.,2021) [ 32 ] ResNet50, AlexNet, and MobileNet Customize dataset has 45,000 images [ 33 ] Accuracy: 98.2% (Hiten G. et al.,2022) [ 12 ] CNN architecture Kaggle data set (4000 images) Accuracy: 98% (Tomás J. et al.,2021) [ 34 ] CNN, Transfer learning, MobileNetv2 Self-gathered 81.2% Accuracy (Jumana W. et al.,2022) [ 35 ] A customized deep CNN RMFD A Accuracy: 99.57% (Pranjali S.et al.,2022) [ 36 ] SSDMNV2, LeNet-5, AlexNet (Alex K. et al.,2017) [ 37 ], VGG-16 (Karen S.,2015) [ 38 ], ResNet (Kaiming H.,2016) [ 18 ] and SVM A customized mask-face-dataset Accuracy: 98.7% (Sneha. S. et al.,2021) [ 39 ] Deep learning 5000 images with-masks 4000 images without-masks Accuracy: 79.24% (Safa T. et al.,2021) [ 40 ] Deep MaskNet framework-MDMFR 3835 images, 1919 images with masks and 1916 images without masks Accuracy: 93.33 7. Conclusions and Future Works The wide variety of face masks that are available, diverse camera resolutions, varying levels of impediments, and several variants (including small dimensions, posture variation, shadows, lighting conditions, viewing angles, and rotation) make mask detection/recognition even more difficult. This has created some difficulties for systems not trained to handle masked faces. As well as low-resolution images, face expressions, storage capacity, limited processing power, and lack of real-world datasets. Therefore, work aimed to determine whether it is possible to recognize masked-faces with a high degree of accuracy. The suggested method used CNN, Pytorch, Keras, and OpenCV to determine whether or not people were wearing facial-masks. We are building an accurate solution by tuning the hyperparameters to accurately identify the existence of masks without generating overfitting. Furthermore, on a sizable face-mask dataset, the suggested technique produces innovative outcomes, which can contribute to public healthcare and be tested on real-time video streams. This study revealed that the proposed system achieved 99.9% accuracy in detecting masked-faces, depending on an imbalanced, large dataset. It is a noteworthy method for executing fast and efficient face recognition, and for defective masks and faces, the modified system still has the best effect and performance. Finally, the work suggests promising future possibilities for researchers. To begin with, the suggested technique is relevant to mask detection and might be a part of any high-definition video monitoring system. Also, the model may be improved and trained on mask datasets, including various pictures associated with appropriately or wrongly wearing masks, ultimately achieving the goal of identifying facemasks to lower the risk of infectious diseases. Declarations Funding Information This research did not obtain financial support from either public or private sources. Author Contribution Ammar HJ. did the main ideation, the primary draft of the paper, and some experiments. Ammar HJ. Designed the model and the computational framework ,analyzed the data, and participated in writing the manuscript. Ahmed Al-Taie and Amal SA. conducted the implementations of codes. Ahmed Al-Taie and Amal SA. did the experiments and helped with manuscript writing. Ammar HJ. did the supervision and wrote the final draft of the paper. All the revisions are done by Ahmed Al-Taie and Amal SA. References Fatima D, Manar A, Qassim N, Tracy S (2024) Artificial intelligence techniques in financial trading: A systematic literature review. J King Saud Univ - Comput Inform Sci 36(3):102015. https://doi.org/10.1016/j.jksuci.2024.102015 Xiuli C, Joohan R (2024) Artificial Intelligence (AI) and Civilization Evolution: Technology, Strategy, and Societal Transformation. Int J Latest Res Humanit Social Sci (IJLRHSS) 07(2):127–140 Maad M (2023) Deep Convolutional Neural Network Architecture to Detect COVID-19 from Chest X-Ray Images, Iraqi Journal of Science, 2023; 64(5): 2561–2574. https://doi.org/10.24996/ijs.2023.64.5.38 Muna K, Ban N (2022) MSRD-Unet: Multiscale Residual Dilated U-Net for Medical Image Segmentation, Baghdad Science Journal. 2022, 19(6): 1603–1611, https://dx.doi.org/10.21123/bsj.2022.7559 Noor M, Nassir H (2022) Deep Learning-Based Segmentation and Classification Techniques for Brain Tumor MRI: A Review. Journal of Engineering, 28(12):93–112. https://doi.org/10.31026/j.eng.2022.12.07 Alessandra L, Loris N, Gianluca M (2023) Deep learning for plankton and coral classification. Appl Comput Inf 19(3/4):265–283. https://doi.org/10.1016/j.aci.2019.11.004 Suranjana M, Annwesha B, Tanusree S (2023) An Observation and Analysis the role of Convolutional Neural Network towards Lung Cancer Prediction. Baghdad Sci J 20(6):2568–2592. https://bsj.uobaghdad.edu.iq/index.php/BSJ/article/view/9029 Loris N, Stefano G, Shery B (2021) Ensemble of convolutional neural networks for bioimage classification. Appl Comput Inf 17(1):19–35. https://doi.org/10.1016/j.aci.2018.06.002 Chamandeep V, Neeraj S (2022) Face and Face-mask Detection System using VGG-16 Architecture based on Convolutional Neural Network, International Journal of Computer Applications, 183(50):16–21, https://doi.org/10.5120/ijca2022921700 Maminiaina A, Andry H, Sabine H, Rajaonarison F, Mirado M, Hajasoa M (2023) Neural networks contribution in face mask detection to reduce the spread of COVID-19, Multimedia Tools and Applications, 82:32559–32581, https://doi.org/10.1007/s11042-023-14920-1 Alaa E (2023) Face recognition using ensemble statistical local descriptors. J King Saud Univ – Comput Inform Sci 35(9):101729. https://doi.org/10.1016/j.jksuci.2023.101729 Hiten G, Karanveer S, Charanjeet S, Abhilasha J, Swati J (2022) A Real Time Face Mask Detection System Using Convolutional Neural Network. Multimed Tools Appl 81:14999–15015. https://doi.org/10.1007/s11042-022-12166-x Ashraf A, Tawfeeq M, Marwa A (2022) Constructing a Software Tool for Detecting Face Mask-wearing by Machine Learning, Baghdad Science Journal, 19(3): 642–653, https://dx.doi.org/10.21123/bsj.2022.19.3.0642 Wadii B, Ayyub A, Aseel A, Muhanad A, Ibrahim A, Maha D (2021) A Deep Learning-based Approach for Real-time Facemask Detection, Conference, 20th IEEE International Conference on Machine Learning and Applications (ICMLA), https://dx.doi.org/10.1109/ICMLA52953.2021.00238 Mark S, Andrew H, Menglong Z, Andrey Z, Liang-Chieh C (2018) MobileNetV2: Inverted Residuals and Linear Bottlenecks. IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 4510–4520, http://dx.doi.org/10.1109/CVPR.2018.00474 Nagrath P, Jain R, Madan A, Arora R, Kataria P, Hemanth J (2021) SSDMNV2: A real-time DNN-based face mask detection system using single shot multibox detector and MobileNetV2, Sustainable cities and society, 66: 102692, https://doi.org/10.1016%2Fj.scs.2020.102692 Dey S, Howlader A, Deb C Mobilenet mask: A multi-phase face mask detection model to prevent person-to-person transmission of sars-cov-2, in Proc. 3rd Int. Trends in Computational and Cognitive Engineering, Conf. UTHM, Malaysia (2021) Springer, 603–613, http://dx.doi.org/10.1007/978-981-33-4673-4_49 Kaiming H, Xiangyu Z, Shaoqing R, Jian S (2016) Deep residual learning for image recognition, in Proc. Computer Vision and Pattern Recognition (CVPR), IEEE Conf., Las Vegas, USA, 770–778. http://dx.doi.org/10.1109/CVPR.2016.90 Sravani N, Kalyani M, Sivapriya G, Triveni B, Abhishek CH (2022) CNN based Face Mask Detection for COVID19 Safety. J Eng Sci 13(7):503–511 Sagar A, Patra J, Suman K (2022) Convolutional neural network architecture based automatic face mask detection. Int J Health Sci 6(S3):623–629. https://doi.org/10.53730/ijhs.v6nS3.5401 Bose S, Logeswari G, Thavavel V, Tariq A, Fadl D, Fahima H et al (2023) A convolutional neural network for face mask detection in IoT-based smart healthcare systems, Computational Physiology and Medicine, 14(1664), https://doi.org/10.3389%2Ffphys.2023.1143249 Lin TY, Maire M, Belongie S, Hays J, Perona P, Ramanan D, Dollár P, Zitnick CL (2014) Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6–12, 2014, Proceedings, Part V 13 2014 (pp. 740–755). Springer International Publishing Naufal M, Febriyanti S, Kurniawan N (2023) Face Mask Detection Under Low Light Condition Using Convolutional Neural Network (CNN). JIPI 8(1):281–290. http://dx.doi.org/10.29100/jipi.v8i1.3324 Ritesh T, Aniqa S, Muhammad A (2023) Face mask detection model based on deep CNN technique using AWS. ijera 13(5):12–19 Sathwika B, Tejaswini A, Varsha A, Baby Rani N (2023) Robust Face Mask Detection Using Deep Learning CNN: An Application of Covid-19. Turkish J Comput Math Educ 14(1):282–295 Bingshu W, Yong Z, Philip Chen CL (2021) Hybrid transfer learning and broad learning system for wearing mask detection in the covid-19 era. IEEE Trans Instrum Meas. http://dx.doi.org/10.1109/TIM.2021.3069844 Jonathan S, Jorge B, Calvopina P, José V (2021) Facial recognition system for people with and without face mask in times of the covid-19 pandemic, Sustainability,13(12), http://dx.doi.org/10.3390/su13126900 Al-Dmour H, Tareef A, Alkalbani A, Hammouri A, Alrahmani B (2023) Masked Face Detection and Recognition System Based on Deep Learning Algorithms. J Adv Inform Technol 14(2):224–232 Mohamed L, Gunasekaran M, Mohamed H, Nour Eldeen M (2021) A hybrid deep transfer learning model with machine learning methods for face mask detection in the era of the COVID-19 pandemic, journal of the International Measurement Confederation, 1:167:108288, https://doi.org/10.1016/j.measurement.2020.108288 Vaibhav J, Surbhi V (2022) Face mask detection using convolutional neural network, in Proc. 12th Int. Cloud Computing Data Sci. Eng. Conf., Noida, India,26–30 https://doi.org/10.1109/Confluence52989.2022.9734156 Francesco M, Antonella S (2021) Transfer learning for mobile real-time face mask detection and localization, Journal of the American Medical Informatics Association, 28(7): 1548–1554, http://dx.doi.org/10.1093/jamia/ocab052 Sethi S, Kathuria M, Kaushik T (2021) Face mask detection using deep learning: An approach to reduce risk of Coronavirus spread. J Biomed Inform 120:103848. https://doi.org/10.1016/j.jbi.2021.103848 Kaggle (2021) Face data hybrid, Balanced Facemask Dataset. https://www.kaggle.com/mrviswamitrakaushik/facedatahybrid Tomás J, Rego A, Viciano-Tudela S, Lloret J (2021) Incorrect facemask-wearing detection using convolutional neural networks with transfer learning, Healthcare, 9(8):1050, https://doi.org/10.3390/healthcare9081050 Jumana W, Thekra A, Taha M (2022) Facemask Wearing Detection Based on Deep CNN to Control COVID-19 Transmission, in Proc. 2nd Int. Engineering Science and Technology (MICEST), Conf. Samawah, Iraq, 158–161. https://doi.org/10.1109/MICEST54286.2022.9790197 Pranjali S, Amitesh G, S (2022) A Comprehensive Analysis on Masked Face Detection Algorithms. Advanced Healthcare Systems. John Wiley & Sons, Ltd., Hoboken, NJ, USA, DOI. https://doi.org/10.1002/9781119769293.ch16 . Alex K, ilya S, Geoffrey EH (2017) ImageNet Classification with Deep Convolutional Neural Networks. Commun ACM 60(6):84–90. https://dx.doi.org/10.1145/3065386 Karen S, Andrew Z (2015) Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1409.1556 Sneha S, Khushboo S (2021) Face Mask Detection for Covid_19 Pandemic Using Pytorch in Deep Learning. IOP Conf. Ser Mater Sci Eng. https://dx.doi.org/10.1088/1757-899X/1070/1/012061 Safa T, Seifeddine M, Mohamed A, Abdellatif M (2021) Real-Time Implementation of AI-Based Face Mask Detection and Social Distancing Measuring System for COVID-19 Prevention. Sci Program 2:1–21. https://doi.org/10.1155/2021/8340779 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-4251321\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":true,\"archivedVersions\":[],\"articleType\":\"Research Article\",\"associatedPublications\":[],\"authors\":[{\"id\":290044684,\"identity\":\"78afb7e7-f3ec-4785-bd81-89b98a0f1bfe\",\"order_by\":0,\"name\":\"Ammar Hussein Jassim\",\"email\":\"data:image/png;base64,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\",\"orcid\":\"\",\"institution\":\"University of Baghdad\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Ammar\",\"middleName\":\"Hussein\",\"lastName\":\"Jassim\",\"suffix\":\"\"},{\"id\":290044685,\"identity\":\"c72b41d9-ec9d-487a-8084-d13983345625\",\"order_by\":1,\"name\":\"Ahmed Altaie\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Baghdad\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Ahmed\",\"middleName\":\"\",\"lastName\":\"Altaie\",\"suffix\":\"\"},{\"id\":290044686,\"identity\":\"f94aef25-a0f9-4cc1-afa2-e3c9519f56c9\",\"order_by\":2,\"name\":\"Amal Sufiuh Ajrash\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"University of Baghdad\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Amal\",\"middleName\":\"Sufiuh\",\"lastName\":\"Ajrash\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2024-04-11 08:58:11\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-4251321/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-4251321/v1\",\"draftVersion\":[],\"editorialEvents\":[],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":55003691,\"identity\":\"3fac1c4c-7189-4be9-b1da-5cf8b6e28ee9\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:43:24\",\"extension\":\"jpg\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":56557,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eSample images without mask.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"1.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/68ac5409dc92c2e389d248e4.jpg\"},{\"id\":55000420,\"identity\":\"0bb81d60-dacb-4db4-988c-8d891431a26a\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:24\",\"extension\":\"jpg\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":46218,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eSample images with mask.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"2.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/9d7581cd85421c11b137cdd7.jpg\"},{\"id\":55000421,\"identity\":\"527a3998-4dac-4aeb-a6cd-b1ea3cc3b58b\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:24\",\"extension\":\"jpg\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":25213,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eDataset classes distribution.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"3.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/e3c9c0b126b233d14d089218.jpg\"},{\"id\":55005685,\"identity\":\"d87d4534-6b04-48a9-8045-c225776a754f\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:51:24\",\"extension\":\"jpg\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":21390,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eDataset classes percentage.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"4.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/8ba4d075703877ef44c05c99.jpg\"},{\"id\":55000424,\"identity\":\"75672928-de6c-4f81-94eb-908162103b9e\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:24\",\"extension\":\"jpg\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":55580,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eBlock diagram of face detection and extraction model.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"5.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/4394a41503bd68f1ca5b5767.jpg\"},{\"id\":55000427,\"identity\":\"93b348a4-b080-44a7-80d5-8f8cc92eb4ca\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:24\",\"extension\":\"jpg\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":110153,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eThe proposed architecture of CNN.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"6.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/f8c9df9db322fd8e9570278c.jpg\"},{\"id\":55000433,\"identity\":\"b79549df-4507-454a-9b81-23ef1eb6b7b2\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:24\",\"extension\":\"jpg\",\"order_by\":7,\"title\":\"Figure 7\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":104836,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eThe proposed CNN architecture.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"7.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/9f0cbbaa8da02d69b797b805.jpg\"},{\"id\":55000436,\"identity\":\"ae411cfb-d2dd-4e9e-86cb-7689992fe275\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:25\",\"extension\":\"jpg\",\"order_by\":8,\"title\":\"Figure 8\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":60334,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eTraining and validation accuracy.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"8.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/f39cefd66124534d3ce7ed6c.jpg\"},{\"id\":55000434,\"identity\":\"0fcdd10c-9151-46d5-9fd7-b33346f3952e\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:25\",\"extension\":\"jpg\",\"order_by\":9,\"title\":\"Figure 9\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":28204,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eConfusion matrix.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"9.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/80b65bc81ca59777389baf2b.jpg\"},{\"id\":55000428,\"identity\":\"d1057483-435c-4ea9-91d4-f34399d72e4b\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:24\",\"extension\":\"jpg\",\"order_by\":10,\"title\":\"Figure 10\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":30888,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eReceiver Operating Characteristic (ROC) Curve.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"10.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/3b144b6b85fcd55eea3b8684.jpg\"},{\"id\":55000425,\"identity\":\"f89e4ff5-ab80-4317-83ad-9eb423270197\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:35:24\",\"extension\":\"jpg\",\"order_by\":11,\"title\":\"Figure 11\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":32619,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003ePrecision-Recall Curve (PRC).\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"11.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/55ec84ad61918e8f2d7b3806.jpg\"},{\"id\":55003694,\"identity\":\"28774675-7ed3-4eb1-8aa5-962e34519592\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:43:24\",\"extension\":\"jpg\",\"order_by\":12,\"title\":\"Figure 12\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":158898,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eSome evaluation samples.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"12.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/a5bc93ea1f33b69819748d4e.jpg\"},{\"id\":55003693,\"identity\":\"249257ba-ade1-4389-93cf-1b766ca2b795\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:43:24\",\"extension\":\"jpg\",\"order_by\":13,\"title\":\"Figure 13\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":135553,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eVarious samples from a testing crowd image.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"13.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/20297d321c1ca7a47320426e.jpg\"},{\"id\":55003699,\"identity\":\"f85ad774-c79e-434c-86ec-116c32d11604\",\"added_by\":\"auto\",\"created_at\":\"2024-04-19 18:43:25\",\"extension\":\"jpg\",\"order_by\":14,\"title\":\"Figure 14\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":77382,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eTesting samples with accuracy.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"14.jpg\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/6b18db771694fd621ec3544e.jpg\"},{\"id\":55448709,\"identity\":\"383691c0-36c5-4464-a775-428eacf0f036\",\"added_by\":\"auto\",\"created_at\":\"2024-04-28 05:36:50\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":1302838,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-4251321/v1/ae63d836-64ef-463f-93a9-a06ea3581ed8.pdf\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Improved CNN System for Face Mask Recognition\",\"fulltext\":[{\"header\":\"1. Introduction\",\"content\":\"\\u003cp\\u003eArtificial intelligence (AI) is an everyday occurrence due to technological innovations. AI is being used in various applications, from financial markets [\\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e] virtual assistants driven by AI to self-driving automobiles [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e]. In some areas, like computer vision, the growth of artificial intelligence is extremely astonishing. It allows machines to observe items in a way humans do [\\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e], allowing them to identify, evaluate, and categorize objects. This is made possible using CNN, a dependable methodology that generates novel prospects for computer vision applications [\\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e4\\u003c/span\\u003e].\\u003c/p\\u003e \\u003cp\\u003eAn algorithm for deep learning can assess images and accurately recognize and classify objects in them [\\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e5\\u003c/span\\u003e]. CNN is built in a way similar to how the brain works, and it can do things like analyze pixel data.\\u003c/p\\u003e \\u003cp\\u003eResearchers were significantly assisted by deep learning computational methods to extract important features that best depict the issue [\\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e6\\u003c/span\\u003e]. Indeed, in several domains, including image classification, neural network techniques have been successfully applied, which combine technologies such as speech, face recognition, self-driving cars, cancer detection [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e8\\u003c/span\\u003e], etc. Deep learning demonstrated its effectiveness in detecting various object types for all of these applications. For the task of detecting the presence of masks, a lot of techniques have been developed from scratch [\\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e9\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e10\\u003c/span\\u003e], like the proposed model that will be introduced in these papers. Consequently, face and mask detection systems based on artificial intelligence are growing in popularity now [\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e]. This study proposes developing a system capable of promptly detecting the presence or absence of masks on faces by employ the CNN net architecture to extract distinctive features from the images.\\u003c/p\\u003e \\u003cp\\u003eThe proposed work aims to make these contributions:\\u003c/p\\u003e \\u003cp\\u003e \\u003cul\\u003e \\u003cli\\u003e \\u003cp\\u003eDesign an efficient face mask classifier that detects any face present within the image regardless of its alignment and improves the accuracy and response time consumption.\\u003c/p\\u003e \\u003c/li\\u003e \\u003cli\\u003e \\u003cp\\u003eAdapt the model with high performance on high and small-size masked faces with different resolutions, formats, and blurry images.\\u003c/p\\u003e \\u003c/li\\u003e \\u003cli\\u003e \\u003cp\\u003eA model that can utilize CPU cores and hyperparameters tuner advantages.\\u003c/p\\u003e \\u003c/li\\u003e \\u003cli\\u003e \\u003cp\\u003eA sizable dataset is required to complete this challenge and train deep-learning models to recognize whether people wear masks.\\u003c/p\\u003e \\u003c/li\\u003e \\u003c/ul\\u003e \\u003c/p\\u003e\"},{\"header\":\"2. Related Work\",\"content\":\"\\u003cp\\u003eThis section includes the most recent and pertinent academic research on face-mask identification using CNN models.\\u003c/p\\u003e \\u003cp\\u003eDozens of journal papers were obtained, and all the papers [\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e] to [\\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e25\\u003c/span\\u003e] were selected for related review because of their detailed explanations, excellent experiments, and good discussions. Some papers are not considered because they don't have any experiments, especially ones without quantitative results; visual detection results aren't shown; or the dataset doesn't have enough images, say 1000 or more. Certain writing works, in particular, use similar methods and only try algorithms on different data sets. As a result, only those with large datasets and good performance were chosen for this work.\\u003c/p\\u003e \\u003cp\\u003eWadii B. et al., 2021 [\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e] recommend using MobileNetV2 [\\u003cspan citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e] for facemask recognition in real-time. Several experiments show the performance of the suggested model (test and training accuracy is 99%). The dataset contains 21407 images. For training, use 17125 images; for testing, use 4282 images.\\u003c/p\\u003e \\u003cp\\u003eNagrath et al., 2021 [\\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e16\\u003c/span\\u003e] proposed SSDMNV2 as a model that uses a single-shot multi-box (SSD) and a MobilenetV2 architecture for face detection, which serves as the classifier's framework. To create a balanced dataset, it uses a dataset with 5521 images with \\u0026ldquo;with_mask\\u0026rdquo; labels and 5521 images without \\u0026ldquo;without_mask\\u0026rdquo; labels. The suggested model has an F1 score of 9.3% and an average accuracy of 92.64%.\\u003c/p\\u003e \\u003cp\\u003eDey S. et al., 2021 [\\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e17\\u003c/span\\u003e] used MobileNetV2 as a classifier, which relies on how well ResNet-10 [\\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e18\\u003c/span\\u003e] and SSD find ROI. They used two distinct datasets. The first dataset had 3835 samples from two classes (1916 with mask and 1919 without mask), and dataset 2, consisting of 1376 images, was used to assess the model's detection capabilities from images and a video stream. Experiment results indicate that with 20% validation samples, MobileNetV2 achieves a 93% accuracy rate for dataset 1, whereas it achieves an accuracy close to 100% for dataset 2.\\u003c/p\\u003e \\u003cp\\u003eSravani N. et al., 2022 [\\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e19\\u003c/span\\u003e], Authors proposed the MobileNet Mask system, a multi-phase recognition technique based on deep learning. The results show that MobileNet achieves around 93% accuracy using 770 validation images and approximately 100% accuracy with 276 validation samples.\\u003c/p\\u003e \\u003cp\\u003eSagar A. et al., 2022 [\\u003cspan citationid=\\\"CR20\\\" class=\\\"CitationRef\\\"\\u003e20\\u003c/span\\u003e], Authors describe a system using OpenCV, Keras packages, and MobileNetV2. 70% of the dataset was used for training and 30% for testing. The obtained accuracy of the system is 96.07%.\\u003c/p\\u003e \\u003cp\\u003eBose S. et al., 2023 [\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e] The proposed system uses YOLOv3 architecture and computer vision. This system is evaluated using the COCO dataset [\\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e] containing 5,000 face-masked and unmasked pictures. Testing data is 20%, and training data is 80%. Experimental results show that the proposed system predicts mask-wearers with 99.2% accuracy and an F1-score of 0.99.\\u003c/p\\u003e \\u003cp\\u003eNaufal M. A. et al., 2023 [\\u003cspan citationid=\\\"CR23\\\" class=\\\"CitationRef\\\"\\u003e23\\u003c/span\\u003e] This study demonstrated that a CNN model with OpenCV, Keras, and TensorFlow can be trained to detect facial masks in low illumination conditions using a recreational low-light image, producing an accuracy of 97.89%. Divide the data into training and test sets with a distribution ratio 70:30. consist of 714 with_mask images and 708 without_mask images; they also include 24 low-illumination images of with_mask and 22 of without_mask.\\u003c/p\\u003e \\u003cp\\u003eRitesh T. et al., 2023 [\\u003cspan citationid=\\\"CR24\\\" class=\\\"CitationRef\\\"\\u003e24\\u003c/span\\u003e] proposed an MTCNN system based on deep learning. Within the 1539 samples that comprise the dataset, 10% are used for validation, while 90% are for training. The suggested dataset was divided into with and without mask images, and 308 images were used for testing the system. The suggested model classifies 172 as masked images and 136 as without mask images, with 99.76% training and 99.35% testing accuracy, respectively.\\u003c/p\\u003e \\u003cp\\u003eSathwika B. et al., 2023 [\\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e25\\u003c/span\\u003e] suggested ResNet-CNN, and basic machine learning tools like OpenCV, TensorFlow, Keras, and Scikit-Learn were used. There are 1376 images that CNN used to train its system, and it can make 98% accurate predictions.\\u003c/p\\u003e\"},{\"header\":\"3. Dataset Description \\u0026 Preparation\",\"content\":\"\\u003cp\\u003eA sizable dataset is required for this purpose to train deep learning models to distinguish between people with masks and those who are not. In this work, we collect our datasets from multiple sources. Indeed, our datasets are made up of a mix of freely available datasets and data that we got from the Google website. The idea was to make it diverse and unbiased. The images are various sizes, the faces are real and sketches, and the pictures are taken at different angles. It includes the complex images of sophisticated face masks printed with printed patterns, textures, or brands. Also, the image has faces that are partially obscured, such as by a beard, hat, hair, hijab, sunglasses, and hands covering the face. Figure\\u0026nbsp;1. shown some images of non-mask-wearing individuals while Fig.\\u0026nbsp;2. shown people using face masks.\\u003c/p\\u003e\\n\\u003cp\\u003eThe dataset contains two folders containing images in jpg format. There are 85,864 images belonging to 2 classes: 46.735 with masks and 39.129 without masks. For evaluation, 75% can be primarily used to train face mask classifiers and 25% for testing by randomly selecting images from the dataset.\\u003c/p\\u003e\\n\\u003cp\\u003eFigure 3. shows a distribution of the dataset. Likewise, Figure 4. presents the percentage of each class\\u0026rsquo;s size.\\u003c/p\\u003e\"},{\"header\":\"4. Proposed Model Architecture\",\"content\":\"\\u003cp\\u003eThe proposed work employs deep neural networks, which are more effective than alternative classification algorithms [\\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e6\\u003c/span\\u003e]. However, training a deep neural network is extremely expensive due to its computationally intensive and time-consuming nature. Therefore, the proposed model utilizes the Num worker (processes) in CPU cores for data loading and GPU capabilities.\\u003c/p\\u003e \\u003cp\\u003eIn a sense, the CPU loads data more efficiently if the Num workers are greater than 1. Therefore, the GPU must wait less, especially when the large dataset is put in different folders on different drives. Therefore, the proposed model used two workers when preprocessing and augmenting the images.\\u003c/p\\u003e \\u003cp\\u003eWhen preparing a dataset, input images from this dataset may need to be preprocessed and augmented. For this purpose, the proposed model, instead of processing input images one by one, replicates the pre-processing and augmentation tasks into units that can be executed concurrently, batching them together for parallel processing (multi-threading). Pre-processes a group of data from the dataset. As a result of the fact that input images are not dependent on one another, each thread can handle a subset of the input images, enabling operations to be parallelized across multiple cores of the CPU. This can help exploit the parallelism offered by modern CPUs and optimize the augmentation and pre-processing pipeline.\\u003c/p\\u003e \\u003cp\\u003eAfter preparing the dataset, design the proposed system, which includes two parts. The first one is to detect faces and specify the ROI of detected faces for extraction. The second part is a CNN, which considers the main part of the system structure to extract features of faces and convert them into a feature map for classification. The affine transformation is used to detect facial features because there may be changes in face size and orientation in the clipped ROI. The subsequent subsections include a comprehensive depiction of each activity within the proposed architecture:\\u003c/p\\u003e \\u003cdiv id=\\\"Sec5\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e4.1 Image Preprocessing and Augmentation\\u003c/h2\\u003e \\u003cp\\u003eWithin the context of this research, a series of steps are involved in preparing the images for training tasks such as normalization, resizing, and augmentation techniques such as rescale, height shift, width shift, rotation, zoom range, and horizontal flip.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec6\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e4.2 Haar Cascade classifier\\u003c/h2\\u003e \\u003cp\\u003eThe input image may vary in size, have multiple faces in a single image, and have non-frontal faces. The initial phase of our approach is to identify individual faces from one image by implements a haar cascade classifier with pre-trained XML files in OpenCV Python to detect faces. Figure\\u0026nbsp;\\u003cspan refid=\\\"Fig5\\\" class=\\\"InternalRef\\\"\\u003e5\\u003c/span\\u003e. shows the first part of the proposed system.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec7\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e4.3 Convolutional Layer\\u003c/h2\\u003e \\u003cp\\u003eFirstly, employ a CNN with an initial hyperparameter configuration and input images processed through a sequence of operations. The precise structure of the proposed networks is shown in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003e. The CNN output categorizes the faces as either having a mask or not.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eThe suggested model has five convolutional layers, two layers are fully connected, and one dense Softmax classifier at the last layer. The 3x3 kernel sizes used in each convolutional layer were placed on top.\\u003c/p\\u003e \\u003cp\\u003eThe convolutional layers comprise 16, 32, 64, and 128, and the system utilizes 256 kernel filters, each having a size of 3x3. As the image goes through the convolutional layers, it changes from an RGB image with 3 depth levels to 96x96x3. The output obtained from each convolution layer is subsequently sent to the max pooling layer with a stride of 2, where pooling is performed using a (2x2) window.\\u003c/p\\u003e \\u003cp\\u003eThe last levels of convolution layers are completely linked hidden layers (flattened layer) of 2304 units, followed by fully connected dense layers containing 1024 units and a dense softmax output layer of 2 units. To prevent overfitting, the 0.2 dropout is used to drop layers at random.\\u003c/p\\u003e \\u003cdiv id=\\\"Sec8\\\" class=\\\"Section3\\\"\\u003e \\u003ch2\\u003e4.3.1 Activation Function\\u003c/h2\\u003e \\u003cp\\u003eIn the proposed CNN, Rectified Linear Unit (ReLU) activation is applied following the convolutional layers, and softmax is a decider in the last layer.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec9\\\" class=\\\"Section3\\\"\\u003e \\u003ch2\\u003e4.3.2 Pooling Layer\\u003c/h2\\u003e \\u003cp\\u003ePooling works on the concept of down-sampling. It reduces the complexity and dimensionality of further layers and eliminates noisy activations. MaxPooling with 2x2 filters and stride 2 is the most common type, and it is the one that is applied in the proposed model.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec10\\\" class=\\\"Section3\\\"\\u003e \\u003ch2\\u003e4.3.3 Hyperparameter Optimization\\u003c/h2\\u003e \\u003cp\\u003eOur goal should be to find the best set of hyperparameter values to get the perfect prediction results from our model. Thus, the suggested model's performance is evaluated on a separate validation dataset to prevent overfitting and select the best hyperparameters depending on the score of the validation results in an iterative process. So, it's important to consider the specific characteristics of the used dataset and problem domain when optimizing the hyperparameters.\\u003c/p\\u003e \\u003cp\\u003eThe proposed model uses Python automated libraries (GridSearchCV in the Scikit API library) for systematic hyperparameter optimization. This method examines a range of hyperparameter combinations (learning rate, batch size, epochs, and dropout) one by one, passes them to the model, and evaluates their performance to find the optimal set that maximizes the model's performance.\\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec11\\\" class=\\\"Section3\\\"\\u003e \\u003ch2\\u003e4.3.4 CNN Model Visualization\\u003c/h2\\u003e \\u003cp\\u003eThe modified model is shown in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003e. takes an input of shapes 96,96,3 from the previous stage and, at the output, classifies each face in the image as a mask or without a mask image.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e \\u003c/div\\u003e\"},{\"header\":\"5. Result And Analysis\",\"content\":\"\\u003cp\\u003eIn this work, the proposed dataset contains 39129 images with mask faces and 46735 images without mask faces, divided into training and test sets. There is an imbalance of classes in the training data. For testing purposes, the dataset contains 21466 images, 11638 instances of \\\"mask,\\\" and 9828 instances of \\\"without_mask.\\\" Evaluating and predicting real-world images was proposed to ensure the modified model's effectiveness.\\u003c/p\\u003e \\u003cp\\u003eOnce the classification model has been built and fitted to the training data, it is necessary to evaluate the model's performance and specify how well it generalizes to the tested data by making predictions on the test data using the prediction method. Therefore, the focus is mainly on classification issues by reporting different metrics for each class in the dataset. For this purpose, a classification report offers valuable information about the model\\u0026rsquo;s performance for each class. Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e Describe the performance of the model in detail.\\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\\u003eClassification report.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"5\\\"\\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=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eClass\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003ePrecision\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eRecall\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eF1-score\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003esupport\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eMask\\u003c/b\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.9796\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.9720\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.9758\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e11638\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eWithout_mask\\u003c/b\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.9671\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.9761\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.9716\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e9828\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eaccuracy\\u003c/b\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colspan=\\\"2\\\" nameend=\\\"c3\\\" namest=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.9739\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e21466\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eMacro avg\\u003c/b\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.9734\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.9740\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.9737\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e21466\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eWeighted avg\\u003c/b\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.9739\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.9739\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.9739\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e21466\\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\\u003eOur model was trained on 35 epochs, and as seen from Table\\u0026nbsp;\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e, we got a training accuracy of 98.32% with a training loss of 5.06% and a testing accuracy of 97.39% with a test loss of 17.50%, meaning that 97% of the predictions were correct. The obtained accuracy is not relative to a particular class, but performance over all classes, and a low mean squared error was also calculated to be equal to 0.26, which proves the proposed classifier is closer to forecasting the true value, which is quite good considering the size and complexity of the dataset and the good design of the proposed system architecture.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab2\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 2\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003ePerformance evaluation for each epoch.\\u003c/p\\u003e \\u003c/div\\u003e \\u003c/caption\\u003e \\u003ccolgroup cols=\\\"5\\\"\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e \\u003cdiv align=\\\"char\\\" char=\\\".\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e \\u003cthead\\u003e \\u003ctr\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003eEpoch\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eLoss\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eAccuracy\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eVal_loss\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003eVal_accuracy\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e0\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.563913\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.738605\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.290697\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.876087\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e1\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.289507\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.877378\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.215679\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.913276\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003cp\\u003e.\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e30\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.051584\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.981471\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.121372\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.972904\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e31\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.058623\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.981617\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.139657\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.970730\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e32\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.050712\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.982374\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.131059\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.971739\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e33\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.55145\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.983135\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0130244\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.973758\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e34\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003e0.050596\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e0.983287\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e0.130150\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"char\\\" char=\\\".\\\" colname=\\\"c5\\\"\\u003e \\u003cp\\u003e0.974730\\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\\u003eFurthermore, as Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig8\\\" class=\\\"InternalRef\\\"\\u003e8\\u003c/span\\u003e. shows, our training history plot is smooth, Confirming the absence of overfitting within the training phase.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eThe tested proposed model has no overlap of results and can discriminate between images with and without a mask with precision, recall, and an F1-score of around 97% for both classes, indicating balanced performance across both classes. Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e,\\u003cspan refid=\\\"Tab2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003e shows the different scores that we calculated for the improved model.\\u003c/p\\u003e \\u003cp\\u003eThe accuracy of positive predictions for the \\u0026ldquo;Mask\\u0026rdquo; class scored 0.9796, indicating that \\u0026asymp;\\u0026thinsp;98% of the masked images predicted as \\u0026ldquo;Mask\\u0026rdquo; are correct. Regarding the class \\\"Without_mask,\\\" the precision is 0.9671, indicating that \\u0026asymp;\\u0026thinsp;97% of the unmasked images predicted as \\u0026ldquo;Without_mask\\u0026rdquo; are correct. Clearly, the modified classifier returns decent results with high true positives.\\u003c/p\\u003e \\u003cp\\u003eIn contrast, the recall is high, equal to around 97%, which indicates the model returns most positive instances, minimizing false negatives but making fewer false positive errors, with some false positives as accrued in the confusion matrix in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig9\\\" class=\\\"InternalRef\\\"\\u003e9\\u003c/span\\u003e.\\u003c/p\\u003e \\u003cp\\u003eThe \\\"Mask\\\" class has an F1-score of 0.9758, balancing precision and recall. Regarding the \\\"Without_mask\\\" category, the F1-score is 0.9716. This value indicates that the enhanced model performs well at predicting whether or not faces will wear masks. Therefore, a higher F1 score indicates a better overall performance.\\u003c/p\\u003e \\u003cp\\u003eThe classification report accuracy, considering the class distribution equal to 0.9734, is a macro-average of the accuracies for each class rather than calculating overall accuracy, which can be more informative in the case of imbalanced classes.\\u003c/p\\u003e \\u003cp\\u003eFigure \\u003cspan refid=\\\"Fig9\\\" class=\\\"InternalRef\\\"\\u003e9\\u003c/span\\u003e. shows the confusion matrix we constructed to measure performance measures. The model can accurately categorize 11,312 out of the 11,638 test pictures using a mask. This is also visible by the high sensitivity value of class \\\"mask\\\"\\u0026mdash;just 97%. Basically, the suggested model classifies most of the \\u0026ldquo;mask\\u0026rdquo; and \\u0026lsquo;without_mask\\u0026rsquo; images correctly at an acceptable performance.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eAnother diagnostic tool commonly used is the region under the arc and the precision-recall curve. As seen in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig10\\\" class=\\\"InternalRef\\\"\\u003e10\\u003c/span\\u003e, the curve bows towards the true positive rate (upper left corner (1,1)), which means the model has a good performance for predicting the positive classes.\\u003c/p\\u003e \\u003cp\\u003eFigure\\u003cspan refid=\\\"Fig10\\\" class=\\\"InternalRef\\\"\\u003e10\\u003c/span\\u003e shows the AUC that signifies the area below the ROC curve, which falls between 0 and 1. It measures the entire performance of the proposed model. The obtained AUC score indicates improved model performance, which has the lowest false positive rates (FPR) and the highest true positive rates (TPR). Also, shows the perfect test of the modified model, passing through the left upper corner. It gives the impression of precision, with a recall value of approximately 99%.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eThe PR curve in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig11\\\" class=\\\"InternalRef\\\"\\u003e11\\u003c/span\\u003e. shows the performance of the proposed model (purple) and allows for a quick assessment of good prediction performance. It gives an impression of the tradeoff between precision and recall. The contrast between the obtained precision and recall is very small. As seen in the upper right corner, the curve is getting close to optimal performance.\\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003eThe good test has a PR curve that bows to the right upper corner (where 100% precision and 100% recall are at the peak point (1,1)). These curves give us the shape we would expect for perfect performance. A high space under the curve shows both high recall and precision. The threshold at point (0, 0) is set to 1.0. This indicates that the model does not discriminate between images with and without masks. An ideal classifier with good precision is associated with a low percentage of FPR, and recall relates to a low percentage of FNR.\\u003c/p\\u003e \\u003cp\\u003eFor further comparisons, the kappa and MCC (Matthews correlation coefficient) are taken into consideration as evaluation metrics. The model achieved a 0.947395 kappa result, representing a good agreement between the classification of observed classes and expected classes in the proposed dataset. The obtained value of the MCC is 0.947429, which indicates the proposed model can predict a high percentage of true positives and true negatives. Finally, the performance values obtained more accurately represent the true ability of the classifier to distinguish between the two classes.\\u003c/p\\u003e\"},{\"header\":\"6. Output\",\"content\":\"\\u003cdiv id=\\\"Sec14\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e6.1 Evaluation images\\u003c/h2\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec15\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e6.2 Testing \\u003cem\\u003eimages\\u003c/em\\u003e\\u003c/h2\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003cp\\u003e \\u003c/p\\u003e \\u003c/div\\u003e \\u003cdiv id=\\\"Sec16\\\" class=\\\"Section2\\\"\\u003e \\u003ch2\\u003e6.3 Comparison with State-of-the-Art Detectors\\u003c/h2\\u003e \\u003cp\\u003eThe comparison results among the proposed model and other recent DL face mask detectors are shown in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003e summarizes the experiment\\u0026rsquo;s findings and indicates that the suggested model had significantly higher accuracy and outperformed previous techniques in terms of accuracy. All the models were trained and tested on different datasets, environments, and network parameters.\\u003c/p\\u003e \\u003cp\\u003e \\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab3\\\" border=\\\"1\\\"\\u003e \\u003ccaption language=\\\"En\\\"\\u003e \\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 3\\u003c/div\\u003e \\u003cdiv class=\\\"CaptionContent\\\"\\u003e \\u003cp\\u003eComparative analysis of various models.\\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\\u003eLiteratures\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eNetwork\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eDataset\\u003c/p\\u003e \\u003c/th\\u003e \\u003cth align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eResults\\u003c/p\\u003e \\u003c/th\\u003e \\u003c/tr\\u003e \\u003c/thead\\u003e \\u003ctbody\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003eProposed method\\u003c/b\\u003e\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eDCNN\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eCollected 85,864 images\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e97.47 training 97.39 testing accuracy\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Nagrath P. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR16\\\" class=\\\"CitationRef\\\"\\u003e16\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eDNN, MobilenetV2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e690 Masked\\u003c/p\\u003e \\u003cp\\u003e686 Not-Masked\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e92.64\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Dey S. et al., 2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e17\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eSSD and ResNet-10, MobileNetV2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e3835 real images (IDS1),\\u003c/p\\u003e \\u003cp\\u003e1376 simulated images\\u003c/p\\u003e \\u003cp\\u003e(IDS2)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eIDS1 Accuracy\\u0026thinsp;=\\u0026thinsp;93%, IDS2\\u003c/p\\u003e \\u003cp\\u003eAccuracy\\u0026thinsp;=\\u0026thinsp;100%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Bingshu W. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR26\\\" class=\\\"CitationRef\\\"\\u003e26\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eFaster R-CNN, BLS\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eWearing-Mask-Detection\\u003c/p\\u003e \\u003cp\\u003e(WMD)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eRecall\\u0026thinsp;=\\u0026thinsp;93.54%, Precision\\u0026thinsp;=\\u0026thinsp;94.84%, F1\\u0026thinsp;=\\u0026thinsp;94.19%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Jonathan S. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e27\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eMobileNetV2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e13359 images, self-built\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy\\u0026thinsp;=\\u0026thinsp;99:65%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(AL-Dmour H. et al.,2023)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR28\\\" class=\\\"CitationRef\\\"\\u003e28\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eCNN\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eRMFD, CFR, Masked-FaceNet\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy:99.5\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Mohamed L. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR29\\\" class=\\\"CitationRef\\\"\\u003e29\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eResNet-50, YOLO5\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eMasked Face Dataset (SMFD, RMFD)\\u003c/p\\u003e \\u003cp\\u003e785 Masked 785 Not-Masked\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy:99.49\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Vaibhav J. et al.,2022)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR30\\\" class=\\\"CitationRef\\\"\\u003e30\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eCNN\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eRMFD (Real World Masked Face Dataset) and Kaggle Dataset\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy:96\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Francesco M. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e31\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eMobilenetV2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e2165 Masked\\u003c/p\\u003e \\u003cp\\u003e1930 Not Masked\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy:98.0\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Sethi, S. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR32\\\" class=\\\"CitationRef\\\"\\u003e32\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eResNet50, AlexNet, and MobileNet\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eCustomize dataset has 45,000 images [\\u003cspan citationid=\\\"CR33\\\" class=\\\"CitationRef\\\"\\u003e33\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy: 98.2%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Hiten G. et al.,2022)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eCNN architecture\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eKaggle data set (4000 images)\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy: 98%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Tom\\u0026aacute;s J. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR34\\\" class=\\\"CitationRef\\\"\\u003e34\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eCNN, Transfer learning, MobileNetv2\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eSelf-gathered\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003e81.2% Accuracy\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Jumana W. et al.,2022)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR35\\\" class=\\\"CitationRef\\\"\\u003e35\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eA customized deep CNN\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eRMFD A\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy: 99.57%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Pranjali S.et al.,2022)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR36\\\" class=\\\"CitationRef\\\"\\u003e36\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eSSDMNV2, LeNet-5,\\u003c/p\\u003e \\u003cp\\u003eAlexNet (Alex K. et al.,2017) [\\u003cspan citationid=\\\"CR37\\\" class=\\\"CitationRef\\\"\\u003e37\\u003c/span\\u003e], VGG-16 (Karen S.,2015) [\\u003cspan citationid=\\\"CR38\\\" class=\\\"CitationRef\\\"\\u003e38\\u003c/span\\u003e], ResNet (Kaiming H.,2016) [\\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e18\\u003c/span\\u003e] and SVM\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003eA customized\\u003c/p\\u003e \\u003cp\\u003emask-face-dataset\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy: 98.7%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Sneha. S. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR39\\\" class=\\\"CitationRef\\\"\\u003e39\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eDeep learning\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e5000 images with-masks 4000 images without-masks\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy: 79.24%\\u003c/p\\u003e \\u003c/td\\u003e \\u003c/tr\\u003e \\u003ctr\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e \\u003cp\\u003e\\u003cb\\u003e(Safa T. et al.,2021)\\u003c/b\\u003e [\\u003cspan citationid=\\\"CR40\\\" class=\\\"CitationRef\\\"\\u003e40\\u003c/span\\u003e]\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e \\u003cp\\u003eDeep MaskNet framework-MDMFR\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e \\u003cp\\u003e3835 images,\\u003c/p\\u003e \\u003cp\\u003e1919 images with masks and 1916 images without masks\\u003c/p\\u003e \\u003c/td\\u003e \\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e \\u003cp\\u003eAccuracy: 93.33\\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\":\"7. Conclusions and Future Works\",\"content\":\"\\u003cp\\u003eThe wide variety of face masks that are available, diverse camera resolutions, varying levels of impediments, and several variants (including small dimensions, posture variation, shadows, lighting conditions, viewing angles, and rotation) make mask detection/recognition even more difficult. This has created some difficulties for systems not trained to handle masked faces. As well as low-resolution images, face expressions, storage capacity, limited processing power, and lack of real-world datasets. Therefore, work aimed to determine whether it is possible to recognize masked-faces with a high degree of accuracy.\\u003c/p\\u003e \\u003cp\\u003eThe suggested method used CNN, Pytorch, Keras, and OpenCV to determine whether or not people were wearing facial-masks. We are building an accurate solution by tuning the hyperparameters to accurately identify the existence of masks without generating overfitting. Furthermore, on a sizable face-mask dataset, the suggested technique produces innovative outcomes, which can contribute to public healthcare and be tested on real-time video streams.\\u003c/p\\u003e \\u003cp\\u003eThis study revealed that the proposed system achieved 99.9% accuracy in detecting masked-faces, depending on an imbalanced, large dataset. It is a noteworthy method for executing fast and efficient face recognition, and for defective masks and faces, the modified system still has the best effect and performance.\\u003c/p\\u003e \\u003cp\\u003eFinally, the work suggests promising future possibilities for researchers. To begin with, the suggested technique is relevant to mask detection and might be a part of any high-definition video monitoring system. Also, the model may be improved and trained on mask datasets, including various pictures associated with appropriately or wrongly wearing masks, ultimately achieving the goal of identifying facemasks to lower the risk of infectious diseases.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003ch2\\u003eFunding Information\\u003c/h2\\u003e \\u003cp\\u003eThis research did not obtain financial support from either public or private sources.\\u003c/p\\u003e\\u003ch2\\u003eAuthor Contribution\\u003c/h2\\u003e\\u003cp\\u003eAmmar HJ. did the main ideation, the primary draft of the paper, and some experiments. Ammar HJ. Designed the model and the computational framework ,analyzed the data, and participated in writing the manuscript. Ahmed Al-Taie and Amal SA. conducted the implementations of codes. Ahmed Al-Taie and Amal SA. did the experiments and helped with manuscript writing. Ammar HJ. did the supervision and wrote the final draft of the paper. All the revisions are done by Ahmed Al-Taie and Amal SA.\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\u003cli\\u003e\\u003cspan\\u003eFatima D, Manar A, Qassim N, Tracy S (2024) Artificial intelligence techniques in financial trading: A systematic literature review. J King Saud Univ - Comput Inform Sci 36(3):102015. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.jksuci.2024.102015\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.jksuci.2024.102015\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eXiuli C, Joohan R (2024) Artificial Intelligence (AI) and Civilization Evolution: Technology, Strategy, and Societal Transformation. Int J Latest Res Humanit Social Sci (IJLRHSS) 07(2):127\\u0026ndash;140\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMaad M (2023) Deep Convolutional Neural Network Architecture to Detect COVID-19 from Chest X-Ray Images, Iraqi Journal of Science, 2023; 64(5): 2561\\u0026ndash;2574. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.24996/ijs.2023.64.5.38\\u003c/span\\u003e\\u003cspan address=\\\"10.24996/ijs.2023.64.5.38\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMuna K, Ban N (2022) MSRD-Unet: Multiscale Residual Dilated U-Net for Medical Image Segmentation, Baghdad Science Journal. 2022, 19(6): 1603\\u0026ndash;1611, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://dx.doi.org/10.21123/bsj.2022.7559\\u003c/span\\u003e\\u003cspan address=\\\"10.21123/bsj.2022.7559\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eNoor M, Nassir H (2022) Deep Learning-Based Segmentation and Classification Techniques for Brain Tumor MRI: A Review. Journal of Engineering, 28(12):93\\u0026ndash;112. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.31026/j.eng.2022.12.07\\u003c/span\\u003e\\u003cspan address=\\\"10.31026/j.eng.2022.12.07\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eAlessandra L, Loris N, Gianluca M (2023) Deep learning for plankton and coral classification. Appl Comput Inf 19(3/4):265\\u0026ndash;283. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.aci.2019.11.004\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.aci.2019.11.004\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSuranjana M, Annwesha B, Tanusree S (2023) An Observation and Analysis the role of Convolutional Neural Network towards Lung Cancer Prediction. Baghdad Sci J 20(6):2568\\u0026ndash;2592. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://bsj.uobaghdad.edu.iq/index.php/BSJ/article/view/9029\\u003c/span\\u003e\\u003cspan address=\\\"https://bsj.uobaghdad.edu.iq/index.php/BSJ/article/view/9029\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLoris N, Stefano G, Shery B (2021) Ensemble of convolutional neural networks for bioimage classification. Appl Comput Inf 17(1):19\\u0026ndash;35. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.aci.2018.06.002\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.aci.2018.06.002\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eChamandeep V, Neeraj S (2022) Face and Face-mask Detection System using VGG-16 Architecture based on Convolutional Neural Network, International Journal of Computer Applications, 183(50):16\\u0026ndash;21, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.5120/ijca2022921700\\u003c/span\\u003e\\u003cspan address=\\\"10.5120/ijca2022921700\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMaminiaina A, Andry H, Sabine H, Rajaonarison F, Mirado M, Hajasoa M (2023) Neural networks contribution in face mask detection to reduce the spread of COVID-19, Multimedia Tools and Applications, 82:32559\\u0026ndash;32581, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1007/s11042-023-14920-1\\u003c/span\\u003e\\u003cspan address=\\\"10.1007/s11042-023-14920-1\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eAlaa E (2023) Face recognition using ensemble statistical local descriptors. J King Saud Univ \\u0026ndash; Comput Inform Sci 35(9):101729. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.jksuci.2023.101729\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.jksuci.2023.101729\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eHiten G, Karanveer S, Charanjeet S, Abhilasha J, Swati J (2022) A Real Time Face Mask Detection System Using Convolutional Neural Network. Multimed Tools Appl 81:14999\\u0026ndash;15015. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1007/s11042-022-12166-x\\u003c/span\\u003e\\u003cspan address=\\\"10.1007/s11042-022-12166-x\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eAshraf A, Tawfeeq M, Marwa A (2022) Constructing a Software Tool for Detecting Face Mask-wearing by Machine Learning, Baghdad Science Journal, 19(3): 642\\u0026ndash;653, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://dx.doi.org/10.21123/bsj.2022.19.3.0642\\u003c/span\\u003e\\u003cspan address=\\\"10.21123/bsj.2022.19.3.0642\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eWadii B, Ayyub A, Aseel A, Muhanad A, Ibrahim A, Maha D (2021) A Deep Learning-based Approach for Real-time Facemask Detection, Conference, 20th IEEE International Conference on Machine Learning and Applications (ICMLA), \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://dx.doi.org/10.1109/ICMLA52953.2021.00238\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/ICMLA52953.2021.00238\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMark S, Andrew H, Menglong Z, Andrey Z, Liang-Chieh C (2018) MobileNetV2: Inverted Residuals and Linear Bottlenecks. IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, pp. 4510\\u0026ndash;4520, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttp://dx.doi.org/10.1109/CVPR.2018.00474\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/CVPR.2018.00474\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eNagrath P, Jain R, Madan A, Arora R, Kataria P, Hemanth J (2021) SSDMNV2: A real-time DNN-based face mask detection system using single shot multibox detector and MobileNetV2, Sustainable cities and society, 66: 102692, https://doi.org/10.1016%2Fj.scs.2020.102692\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eDey S, Howlader A, Deb C Mobilenet mask: A multi-phase face mask detection model to prevent person-to-person transmission of sars-cov-2, in Proc. 3rd Int. Trends in Computational and Cognitive Engineering, Conf. UTHM, Malaysia (2021) Springer, 603\\u0026ndash;613, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttp://dx.doi.org/10.1007/978-981-33-4673-4_49\\u003c/span\\u003e\\u003cspan address=\\\"10.1007/978-981-33-4673-4_49\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eKaiming H, Xiangyu Z, Shaoqing R, Jian S (2016) Deep residual learning for image recognition, in Proc. Computer Vision and Pattern Recognition (CVPR), IEEE Conf., Las Vegas, USA, 770\\u0026ndash;778. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttp://dx.doi.org/10.1109/CVPR.2016.90\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/CVPR.2016.90\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSravani N, Kalyani M, Sivapriya G, Triveni B, Abhishek CH (2022) CNN based Face Mask Detection for COVID19 Safety. J Eng Sci 13(7):503\\u0026ndash;511\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSagar A, Patra J, Suman K (2022) Convolutional neural network architecture based automatic face mask detection. Int J Health Sci 6(S3):623\\u0026ndash;629. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.53730/ijhs.v6nS3.5401\\u003c/span\\u003e\\u003cspan address=\\\"10.53730/ijhs.v6nS3.5401\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBose S, Logeswari G, Thavavel V, Tariq A, Fadl D, Fahima H et al (2023) A convolutional neural network for face mask detection in IoT-based smart healthcare systems, Computational Physiology and Medicine, 14(1664), https://doi.org/10.3389%2Ffphys.2023.1143249\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eLin TY, Maire M, Belongie S, Hays J, Perona P, Ramanan D, Doll\\u0026aacute;r P, Zitnick CL (2014) Microsoft coco: Common objects in context. In Computer Vision\\u0026ndash;ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6\\u0026ndash;12, 2014, Proceedings, Part V 13 2014 (pp. 740\\u0026ndash;755). Springer International Publishing\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eNaufal M, Febriyanti S, Kurniawan N (2023) Face Mask Detection Under Low Light Condition Using Convolutional Neural Network (CNN). JIPI 8(1):281\\u0026ndash;290. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttp://dx.doi.org/10.29100/jipi.v8i1.3324\\u003c/span\\u003e\\u003cspan address=\\\"10.29100/jipi.v8i1.3324\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eRitesh T, Aniqa S, Muhammad A (2023) Face mask detection model based on deep CNN technique using AWS. ijera 13(5):12\\u0026ndash;19\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSathwika B, Tejaswini A, Varsha A, Baby Rani N (2023) Robust Face Mask Detection Using Deep Learning CNN: An Application of Covid-19. Turkish J Comput Math Educ 14(1):282\\u0026ndash;295\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eBingshu W, Yong Z, Philip Chen CL (2021) Hybrid transfer learning and broad learning system for wearing mask detection in the covid-19 era. IEEE Trans Instrum Meas. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttp://dx.doi.org/10.1109/TIM.2021.3069844\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/TIM.2021.3069844\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eJonathan S, Jorge B, Calvopina P, Jos\\u0026eacute; V (2021) Facial recognition system for people with and without face mask in times of the covid-19 pandemic, Sustainability,13(12), \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttp://dx.doi.org/10.3390/su13126900\\u003c/span\\u003e\\u003cspan address=\\\"10.3390/su13126900\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eAl-Dmour H, Tareef A, Alkalbani A, Hammouri A, Alrahmani B (2023) Masked Face Detection and Recognition System Based on Deep Learning Algorithms. J Adv Inform Technol 14(2):224\\u0026ndash;232\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eMohamed L, Gunasekaran M, Mohamed H, Nour Eldeen M (2021) A hybrid deep transfer learning model with machine learning methods for face mask detection in the era of the COVID-19 pandemic, journal of the International Measurement Confederation, 1:167:108288, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.measurement.2020.108288\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.measurement.2020.108288\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eVaibhav J, Surbhi V (2022) Face mask detection using convolutional neural network, in Proc. 12th Int. Cloud Computing Data Sci. Eng. Conf., Noida, India,26\\u0026ndash;30 \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1109/Confluence52989.2022.9734156\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/Confluence52989.2022.9734156\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eFrancesco M, Antonella S (2021) Transfer learning for mobile real-time face mask detection and localization, Journal of the American Medical Informatics Association, 28(7): 1548\\u0026ndash;1554, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttp://dx.doi.org/10.1093/jamia/ocab052\\u003c/span\\u003e\\u003cspan address=\\\"10.1093/jamia/ocab052\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSethi S, Kathuria M, Kaushik T (2021) Face mask detection using deep learning: An approach to reduce risk of Coronavirus spread. J Biomed Inform 120:103848. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1016/j.jbi.2021.103848\\u003c/span\\u003e\\u003cspan address=\\\"10.1016/j.jbi.2021.103848\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eKaggle (2021) Face data hybrid, Balanced Facemask Dataset. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.kaggle.com/mrviswamitrakaushik/facedatahybrid\\u003c/span\\u003e\\u003cspan address=\\\"https://www.kaggle.com/mrviswamitrakaushik/facedatahybrid\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eTom\\u0026aacute;s J, Rego A, Viciano-Tudela S, Lloret J (2021) Incorrect facemask-wearing detection using convolutional neural networks with transfer learning, Healthcare, 9(8):1050, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.3390/healthcare9081050\\u003c/span\\u003e\\u003cspan address=\\\"10.3390/healthcare9081050\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eJumana W, Thekra A, Taha M (2022) Facemask Wearing Detection Based on Deep CNN to Control COVID-19 Transmission, in Proc. 2nd Int. Engineering Science and Technology (MICEST), Conf. Samawah, Iraq, 158\\u0026ndash;161. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1109/MICEST54286.2022.9790197\\u003c/span\\u003e\\u003cspan address=\\\"10.1109/MICEST54286.2022.9790197\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003ePranjali S, Amitesh G, S (2022) A Comprehensive Analysis on Masked Face Detection Algorithms. Advanced Healthcare Systems. John Wiley \\u0026amp; Sons, Ltd., Hoboken, NJ, USA, DOI. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1002/9781119769293.ch16\\u003c/span\\u003e\\u003cspan address=\\\"10.1002/9781119769293.ch16\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e.\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eAlex K, ilya S, Geoffrey EH (2017) ImageNet Classification with Deep Convolutional Neural Networks. Commun ACM 60(6):84\\u0026ndash;90. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://dx.doi.org/10.1145/3065386\\u003c/span\\u003e\\u003cspan address=\\\"10.1145/3065386\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eKaren S, Andrew Z (2015) Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv:1409.1556\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSneha S, Khushboo S (2021) Face Mask Detection for Covid_19 Pandemic Using Pytorch in Deep Learning. IOP Conf. Ser Mater Sci Eng. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://dx.doi.org/10.1088/1757-899X/1070/1/012061\\u003c/span\\u003e\\u003cspan address=\\\"10.1088/1757-899X/1070/1/012061\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e \\u003cli\\u003e\\u003cspan\\u003eSafa T, Seifeddine M, Mohamed A, Abdellatif M (2021) Real-Time Implementation of AI-Based Face Mask Detection and Social Distancing Measuring System for COVID-19 Prevention. Sci Program 2:1\\u0026ndash;21. \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://doi.org/10.1155/2021/8340779\\u003c/span\\u003e\\u003cspan address=\\\"10.1155/2021/8340779\\\" targettype=\\\"DOI\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/span\\u003e\\u003c/li\\u003e\\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\":\"Face Mask Detection, Face Mask Recognition, Deep Learning (DL), Convolutional Neural Network (CNN), Machine Learning (ML)\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-4251321/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-4251321/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eDeep learning, especially convolutional neural networks, has significantly improved performance in computer vision. Therefore, we designed and developed a modified deep convolutional neural network framework for detecting mask in facial images in a sizable synthesized and un-synthesized face mask dataset.\\u003c/p\\u003e \\u003cp\\u003eThe suggested method can be utilized to detect face masks in any image with a low-resolution, different alignments, complex, and noisy background by tuning the hyperparameters to accurately identify the existence of masks without generating overfitting.\\u003c/p\\u003e \\u003cp\\u003eThe experimentally obtained results demonstrate that the suggested model exhibits a significant efficiency level, achieving 97.39% accuracy, 97.34% precision, 97.41% recall, 97.37% F1-score, and 97.4% AUC. The empirical results have been documented after 35 iterations using optimized hyperparameter settings, and those predictive models were trained on 64,398 images with a 98% accuracy rate and 0.05 loss, proving the proposed work's reliability and robustness.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Improved CNN System for Face Mask Recognition\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2024-04-19 18:35:19\",\"doi\":\"10.21203/rs.3.rs-4251321/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\":\"0647e908-0990-4a65-a8c0-d63f3227bf62\",\"owner\":[],\"postedDate\":\"April 19th, 2024\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"posted\",\"subjectAreas\":[],\"tags\":[],\"updatedAt\":\"2024-04-28T05:28:35+00:00\",\"versionOfRecord\":[],\"versionCreatedAt\":\"2024-04-19 18:35:19\",\"video\":\"\",\"vorDoi\":\"\",\"vorDoiUrl\":\"\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-4251321\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-4251321\",\"identity\":\"rs-4251321\",\"version\":[\"v1\"]},\"buildId\":\"qtupq5eGEP_6zYnWcrvyt\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}