Research On Worker Unsafe Behavior Detection Algorithm Based On Deep Learning

With the concept of a "smart construction site," the safety topics of construction workers have received widespread attention.The unsafe behavior of construction workers is the main reason for the accident.Therefore,timely supervision and stopping the unsafe behavior of the construction site workers can not only prevent accidents but also improve the completion of the project.In recent years,the continuous development of science and technology has provided technical support for unsafe behavior recognition.Integrating deep learning algorithms with the construction industry has provided efficient technical means for supervising unsafe behavior sites.To this end,this article proposes to detect algorithms based on workers’ unsafe behaviors based on deep learning to detect whether there is unsafe behavior in construction site workers.The main research content is as follows:Firstly,study the detection performance of mainstream detection algorithms on workers’ unsafe behavior data sets.By investigating the types of unsafe behaviors in the construction scene,the testing category is clear,and the data set of unsafe behavior of workers is constructed targeted.Analyze the mainstream object detection model Faster R-CNN,YOLOv4,YOLOv5 l,and YOLOv7’s training detection effect on the workers’ unsafe behavior data set.By comparing the detection accuracy and detection time of different models,finally,select YOLOv7 with better detection performance as the basic network,and conduct targeted improvement and optimization.Secondly,for the problem of poor detection accuracy under small targets and complex background images,a detection algorithm based on improving YOLOv7 is proposed.By adding a 160×160 detection layer to the YOLOv7 network,improve the recognition rate of small targets and achieve multi-scale prediction;introduce the Sim AM attention mechanism,enhance the network’s extraction ability of image features,improve the network’s detection accuracy of complex background goals;At the same time,the SIo U loss function replaces the original network CIo U loss function,introduce the vector angle between the real box and the prediction box,and reduce the total freedom of the loss function to improve the detection accuracy and convergence speed of the model.The above improvement methods verify the construction of the workers’ unsafe behavior data set.The results show that the average accuracy of the improved YOLOv7 algorithm(m AP@0.5)reached 92.8%,and the accuracy of the improved algorithm was significantly improved.Finally,based on the detection algorithm research,the executable graphical user interface(GUI)was developed through the PyQt5 and Pycharm modules,which showed the interdependence of the entire human-machine communication in real-time.After debugging and testing the image interface,it indicates that the designed workers’ unsafe behavior detection interface operation is convenient.It can realize human-machine interaction and save the detection data.It has certain feasibility.