Response to the global pandemic, in this project have developed an innovative mask recognition system using deep learning and computer vision, based on the MobileNetV2 convolutional neural network architecture with a custom classification head. This system effectively distinguishes individuals wearing masks from those without by employing careful pre-processing, data augmentation, and model training. Through meticulous parameter tuning, achieved impressive performance metrics: a training accuracy of 98%, validation accuracy of 97%, and a balanced F1-score of 96%. This well-rounded model strikes a balance between precision and recall, minimizing false positives and false negatives. It\'s a significant contribution to public health and safety, poised to enhance mask compliance and collective well-being in our changing world.
Introduction
I. INTRODUCTION
In the face of persistent global challenges posed by infectious diseases, particularly in pandemic scenarios, the urgency for innovative solutions to safeguard public health and safety has become paramount. This project presents a pioneering approach to automated mask recognition—an essential facet of ensuring public safety compliance in the current context. Central to this endeavour is the deployment of MobileNetV2, a potent architecture of an artificial neural network (CNN) renowned for its efficiency and adeptness in processing visual data. Capitalizing on its intricate design, featuring depth-wise separable convolutions and inverted residual blocks, MobileNetV2 serves as the core feature extractor for discerning intricate patterns within images. Complementing this foundation, a meticulously crafted custom classification head leverages these extracted features to accurately determine mask presence or absence. This tailored approach enhances the model's precision and dependability in mask detection.
Beyond its technical innovations, this project seamlessly integrates modern machine learning methods and tangible real-world concerns, yielding automated solutions applicable across diverse settings. The proposed mechanism for recognizing Masks can transform compliance monitoring by quickly and accurately detecting non-adherence. Guidelines by public health authorities and security personnel. Amid the dynamic landscape of health and safety, this endeavour offers a glimpse into the future of technology-infused public health measures. By harnessing the prowess of MobileNetV2 and custom classification, this mask recognition system not only showcases the prowess of artificial intelligence in addressing societal challenges but also lays the groundwork for intelligent systems prioritizing human well-being.
II. LITERATURE SURVEY
This study concentrated on face mask detection employing a combination of YOLO and CNN models. This unique approach showcased the integration of YOLOv3 with CNN's hidden layers, enabling effective image recognition and localization across various image types. This endeavour gains significance within the context of the global COVID-19 pandemic, necessitating a system of public health services for mask detection This study also delved into vision learning, focusing on accurately tracking small and dense objects' movements in videos using YOLO with their customized dataset. explored different model deep learning, including color-based, fuzzy-based, motion-based, shape-based, and YOLO (You Only Look Once) for object detection.
This study directed their attention towards facial mask recognition and crowd counting, incorporating the RCNN model. Their AI-based system prioritized ethical factors, ensuring data privacy and inclusiveness. It utilized monocular cameras and machine learning-based real-time object sensors to maintain social distancing awareness, triggering warnings without singling out individuals.
In this study a novel SSDMNV2 model for face mask recognition was designed utilizing OpenCV Deep Neural Network (DNN), TensorFlow, and Keras libraries. Employing MobileNetV2 architecture, this model effectively differentiated masked and unmasked frontal faces, assessing mask proper usage.
Data pre-processing, including cleaning and error correction, was vital due to the dearth of top-notch datasets. The object detection algorithm captured multiple objects in scenes using multibox, resembling the YOLO technique.
4. This study introduced an innovative facemask identification method classifying three maskwearing conditions: properly masked, improperly masked, and unmasked. Their approach combined using traditional machine learning for classification and deep transfer learning for feature extraction, improving performance using SVMs, or support vector machines, are conventional classifiers and decision tree. Ensemble techniques were employed to achieve accuracies of 94.54% and 99.49% for the decision tree and SVM classifiers, respectively.
5. This study a ground breaking Face Mask model was crafted, harnessing the capabilities of OpenCV Deep Neural Network (CNN), TensorFlow, and Keras libraries. Leveraging the power of ResNet architecture, this model adeptly discerned individuals with and without masks, scrutinizing mask compliance. Robust data pre-processing, encompassing data cleaning and error rectification, played a pivotal role, given the scarcity of high-quality datasets. The object detection algorithm employed a multi-scale approach, reminiscent of the Faster R-CNN technique, to detect numerous objects within complex scenes.
III. PROPOSED SYSTEM
The proposed system introduces a novel approach that leverages the strengths of both transfer learning and custom classification. The MobileNetV2 CNN serves as a foundational feature extractor, capable of comprehending intricate visual patterns, while a custom classification head tailors the model specifically for mask recognition. This hybrid architecture allows the system to discern minute details that characterize mask usage, enabling accurate and automated classification in the proposed system we used CNN model and MobileV2Net model.
IV. METHODOLOGY
CNNs, or Convolutional Neural Networks, excel in image and video analysis. They autonomously learn and extract visual features, proving invaluable for tasks such as image classification, object detection, and segmentation. CNNs have transformed computer vision, with wide applications in fields like medicine and self-driving cars, thanks to their hierarchical pattern recognition, driving advances in AI and ML. In the process of developing a robust mask recognition system, project has begun by collecting a diverse dataset of individuals wearing and not wearing masks and meticulously pre-processed the images to ensure uniformity. This dataset was then split into training, validation, and testing sets for model evaluation. Leveraging the MobileNetV2 architecture as our foundation and integrated the pre-trained model, omitting its top layers to preserve its feature extraction capabilities. On top of this base, we designed a custom classification head comprising layers such as average pooling, flattening, dense (fully connected) layers, dropout, and softmax activation. The resultant model was compiled with appropriate loss functions, optimizers (utilizing Adam with learning rate scheduling), and evaluation metrics. To enhance model robustness, have implemented data augmentation using the Image Data Generator during training. Throughout the training process, we closely monitored the model's performance on the validation dataset, fine-tuning hyper parameters as necessary to prevent overfitting. After training, we rigorously evaluated the model's performance on the testing dataset, assessing its accuracy, precision, recall, and F1-score, and generated a comprehensive classification report to identify its strengths and weaknesses. Finally, explored the model's deployment in real-world scenarios, testing its accuracy and scalability, and meticulously documented the entire methodology for knowledge dissemination through presentations, reports, and potential publications. Customised MobileNetV2 and CNN model is shown in the Fig No. 1 &2.
Conclusion
Its core aim was to provide a robust solution for accurately detecting mask-wearing individuals, particularly crucial during health crises. The integration of MobileNetV2\'sefficient feature extraction established a strong foundation, while the tailored classification head improved accuracy. Through meticulous training and validation, the system effectively differentiated between masked and unmasked individuals, with broad applications spanning healthcare, education, security, and more. By enforcing mask compliance, this system actively reduces infectious disease transmission, enhancing safety and nurturing secure environments. It harmoniously showcases the potential of advanced machine learning in addressing real-world challenges, exemplifying innovation\'s role in strengthening public health efforts. As society continues navigating health complexities, the automated mask recognition system highlights technology\'s pivotal role. Its success emphasizes AI\'s power to create impactful solutions, underpinned by best practices, cutting- edge architectures, and practical implementation. This accomplishment reinforces the pursuit of healthier, safer, and technologically advanced communities.
References
[1] Mohamed Loey, Gunasekaran Manogaran, Mohamed Hamed N. Taha, Nour Eldeen M. Khalifa, “hybrid deep transfer learning model with machine learning methods for face mask detection in the era of the COVID-19 pandemic,” College of Information and Electrical Engineering, Asia University, Taiwan Measurement 167 (2021) 108288.
[2] Preeti Nagrath, Rachna Jain, Agam Madan, Rohan Arora, Piyush Kataria, Jude Hemanth “SSDM V2: A real time DNN-based face mask detection system using single shot multibox detector and MobileNetV2,” Department of ECE, Karunya Institute of Technology and Sciences, Coimbatore, India Sustainable Cities and Society 66 (2021) 102692.
[3] Mohamed Loey a, Gunasekaran Manogaran, Mohamed Hamed N. Taha, Nour Eldeen M. Khalifa, “Fighting against COVID-19: A novel deep learning model based on YOLO-v2 with ResNet-50 for medical face mask detection,”, University of California, Davis, USA Sustainable Cities and Society 65 (2021) 102600.
[4] Walid Hariri, “Efficient Maske Face Recognition Method During The COVID-19 Pandemic,” Labged Laboratory Department of Computer Science Badji Mokhtar Annaba University December 12, 2020
[5] Md. Sabbir Ejaz, Md. Rabiul Islam, Md Sifatullah, Ananya Sarker, “Implementation of Principal Component Analysis on Masked and Non-masked Face Recognition,” 1st International Conference on Advances in Science, Engineering and Robotics Technology 2019 (ICASERT 2019).