Surface Defect Detection of Industrial Components Based on Improved YOLOv5s

Due to the frequent occurrence of missed detections, false alarms, and the low accuracy of surface defect detection in industrial components, surface defect detection has always been a challenge in the industrial field. In this study, we propose a novel network structure based on YOLOv5s that achieves improved defect detection accuracy by improving the YOLOv5s network structure. First, a CoTNet transformer module is incorporated in the feature extraction process of the backbone network, replacing the original C3 network module. A global attention mechanism (GAM) is integrated along with the C3 module into the neck to enhance the feature learning ability of the model, so that it focuses more on features. Finally, the adaptive spatial feature fusion (ASFF) algorithm is used for prediction to increase the forecasting accuracy of the model. Finally, the improved algorithm is evaluated on a dataset containing NEU-DET steel surface defects. The experimental results show that the improved algorithm in this article outperforms other current state-of-the-art algorithms in terms of this dataset detection accuracy. In addition, we conduct additional experiments on the PUK-Market-PCB dataset, confirming that the proposed method yields improved detection accuracy.