基于特征增强和孪生结构网络的焊缝关键位置检测与跟踪OA

To further improve welding quality,an intelligent detection and tracking system for weld seams based on deep learn-ing is proposed.A detection network for key positions of weld seams is constructed.The network takes convolutional neural network(CNN)as the basic network,includes weld seams feature extraction module,convolutional block attention module(CBAM),feature fusion module,anchor boxes generation and position detection module.The weld seam tracking network based on twin structure is built by using the characteristics of continuity and predictability between welding image frames to im-prove the tracking efficiency for key positions of weld seams.The detection and tracking network is deployed in the system to verify the detection and tracking of weld seams.The results show that the detection precision of the detection constructed net-work is over 90％,with a maximum of 95.27％,and the network loss value is 0.15.The average distance error between posi-tion tracked by the key position tracking network of weld seams and actual position is about 0.100 mm,and the network loss value is 0.07.A system is constructed based on advanced RISC machine(ARM)and field programmable gate array(FPGA),and deep learning algorithm is deployed into the system.The test results show that average error between the tracked position and actual position is 0.001 mm.From this,it can be concluded that the proposed detection and tracking method of weld seams has high tracking precision and can be used for intelligent recognition of weld seams.