Research on Industrial Non-Destructive Testing Technology Based on Improved YOLOv5s

Non-destructive testing technology plays a pivotal role in the manufacturing process of modern industrial products, and is of incredible importance for moving forward the item quality of workpieces. Currently, the commonly used non-destructive testing technology has disadvantages such as complicated operation, high testing cost, and limited scope of application. This paper suggests an upgraded multi-scale target detection algorithm based on YOLOv5s in light of the aforementioned issues. Firstly, an improved channel attention mechanism module ECA-Net is added into the backbone network. Second, the maximum pooling operation in the SPP module is replaced with atrous convolution in this paper. Finally, the channel weight information is added to the concat operation of the PAN structure, which improves the sensitivity of the network to the channel information of the feature map. This paper uses the NEU-DET steel surface defect data set to do a series of experiments to verify the detection effect of the model. According to the experimental results, the mAP of the algorithm proposed in this paper is 73.47%, which is 7.18% higher than that of YOLOv5s, and the FPS is 68.5 frames/s, which satisfies the needs of industrial production environments.