A Study On Improved YOLOv5s Based On Employee Protective Gear Wear Detection For LCD Panel Cutting Lines

Safety supervision and management is an important part of a company’s positive development,with real-time supervision to ensure the safety of the operators themselves and the efficient production of the company,and is even more essential for production lines such as LCD panel cutting,where the dress code for employees is strictly required.At the same time,with the rapid development of computer vision technology,there has been a major breakthrough in the field of target detection in terms of detection speed and hardware requirements.In this thesis,we propose the use of the YOLOv5 s algorithm for target detection of employee protective gear wear in LCD panel cutting lines,in response to traditional safety supervision models.For the problem of low detection accuracy of the original model,an improved CCS-YOLOv5 s algorithm is proposed based on the actual production environment and the initial results of the algorithm.Firstly,the original C3 module is replaced by a CBAM module in the backbone network to guide the model to pay more attention to the feature of protective gear in a complex background with strong and uneven illumination;secondly,the SIo U Loss function is used to replace the CIo U bounding box regression localisation loss function,and an angle penalty term is added to allow the prediction box to locate the target box more quickly and accurately to improve the model recognition accuracy and speed;finally,in view of the Finally,in view of the problems of target occlusion and small sample data set in the actual operation,the Cutout data enhancement technique is applied to expand the sample of occluded data while effectively avoiding the overfitting phenomenon.The experimental results show that compared with the original Yolov5 s algorithm,the m AP value of the CCS-YOLOv5 s model reaches 90.2%,and the average recognition detection of a single image takes only 0.025s;the false detection rate and the missed detection rate of the test set images are reduced by 2.77% and6.52% respectively,and the recall rate and false detection rate on the network image set are 84.31% and 7.32% respectively.The improved model has better accuracy and robustness for strong light uneven and complex occlusion scenes while keeping the number of parameters and detection speed metrics basically unchanged.It also has strong generalization capability for image recognition of different data sources but the same scene,and its overall performance can meet the practical needs of safety supervision of LCD panel cutting production lines.