Research On Wearing Compliance Detection Method Of Chemical Protective Articles Based On Deep Learning

There are many safety accidents in factories and laboratories every day,and many of these safety accidents are caused by workers failing to wear chemical protective equipment.In addition,national safety agencies require supervision of wearing protective equipment throughout the production process.However,there are many problems with manually supervising workers to ensure that they are always wearing protective equipment,and real-time monitoring is not feasible.Object detection based on deep learning can provide a good solution to reduce security risks.Developing and testing these methods requires large amounts of high-quality training data.However,a dataset on chemical plant scenarios and related protective equipment is missing.Moreover,the factory has high requirements for safety monitoring,and it is necessary to achieve accurate and stable real-time monitoring of protective equipment.In order to solve such problems,this paper designs and contributes a dataset based on the factory background and related protective equipment,and produces a series of benchmark experiments for subsequent research.In addition,based on deep learning,two methods for wearing compliance are designed and proposed.The research work in this thesis is as follows:1.Construction of chemical protective equipment dataset(PPED): According to relevant standards,we completed the shooting and production of multiple scenes such as factories,machine and equipment accessories.Using the Chemical Protective Equipment Dataset(PPED),we provide 3,300 photos of people wearing protective eyewear,face shields,gas masks,masks,gowns,and protective gloves.2.Conduct benchmark experiments.On the basis of this dataset,several widely used state-of-the-art object detection models,including YOLOv3-SPP,YOLOv5,SSD,and Faster R-CNN,are used for comparative experiments to create a benchmark for the field.In our study,YOLOv5 achieved the best detection results with 93.6% m AP and 87.5% recall.3.Two wearing compliance detection methods are proposed.The two methods are based on image classification technology and location information detection.The first method is based on image classification technology,through the positioned body parts,the wearing of protective equipment is judged,so as to effectively identify the improper wearing of protective equipment.This method is simple to operate,can improve safety,and is suitable for scenarios where it is necessary to quickly determine the wearing of protective equipment.The second method is based on location information detection technology,through the detected protective equipment and the position information of the corresponding body parts of the human body,accurately identify each person’s wearing situation,and has high real-time and accuracy.This method is suitable for scenarios that require a more granular inspection of the wearing of protective equipment.