Automotive Door Assembly Process Inspection Based on Improved YOLOv5
The assembly work of car doors has achieved automatic assembly line style,but there is currently no ef-fective solution for whether the assembly of door panel parts is in place.A YOLO v5 network optimized automobile door panel assembly inspection network is proposed to address this situation.This network realizes the detection of three types of assembly processes:screws,welding points,and fasteners,and can detect whether they have been correctly installed at the corresponding assembly points.In order to improve the detection accuracy of the as-sembly status of various components,attention mechanism is added to the convolutional module in YOLO v5 net-work to enhance the feature learning of the backbone network for high-frequency backbone networks.Secondly,for the SPPF Receptive field expansion module in the original network,the hole convolution group is used to con-struct Receptive field ranges of different sizes to enrich the feature information,and the maximum pooling layer is used to enhance the high-frequency feature information in the feature map to suppress the interference of back-ground noise.After experimental testing,the optimized network has improved its Precision index by 2.1％to 97.4％,Recall index by 8.4％to 97.0％,and mAP index by 5.9％to 98.1％compared to before optimization,which has certain practicality.
industrial testingautomotive door panelsdeep learningobject detectionYOLO
万方数据
维普
