Research On Safety Helmet Wearing Detection Algorithm Based On Improved YOLOv5

The construction industry is a labor-intensive industry with complex working environment and frequent safety accidents.Due to the complexity of the construction scene,the existing helmet wearing detection algorithm has two problems.On the one hand,the accuracy of small target detection is low,as well as the problems of overlapping,occluding target error detection and missing detection.On the other hand,the algorithm model is difficult to be effectively transplanted to devices with relatively limited computing resources.Therefore,in order to solve the problems encountered by existing helmet wearing detection algorithms,this dissertation has made corresponding improvements on the basis of YOLOv5 algorithm.The main work of this dissertation includes:Firstly,aiming at the problems of small targets,overlapping and occluding targets in the helmet wearing detection scene,a helmet wearing detection algorithm E-YOLOv5 is proposed to improve YOLOv5.The main improvements are as follows:(1)Add an effective feature extraction layer to the backbone network,so that the algorithm can detect small targets more easily.(2)The original feature pyramid structure is improved to enhance feature fusion and improve the accuracy of small target detection.(3)DIOU-NMS is introduced to replace NMS,which solves the problem of false detection and missing detection of overlapping and occluding targets.Secondly,in view of the problem that E-YOLOv5 algorithm model is difficult to be effectively transplanted to devices with relatively limited computing resources,E-YOLOv5 algorithm is further improved to obtain E-YOLOv5 tiny algorithm.The main improvements are as follows:(1)Shuffle Net V2 network is used as the basic backbone network,which makes the network feature extraction process lightweight.(2)The improved attention mechanism module is introduced into the backbone network to ensure the accuracy of detection.The experimental results show that the average average precision(m AP)of E-YOLOv5 algorithm has reached 96.7%,which is 3.3% higher than the original YOLOv5 algorithm,meeting the requirements for the accuracy of helmet wearing detection in the construction scene.E-YOLOv5 tiny lightweight algorithm has smaller model volume and sufficient detection accuracy,which is more suitable for deployment to some devices with limited computing resources.Figure[32] Table[6] Reference[57]...