Abstract
Due to the small defect areas and indistinct features on industrial components, detecting surface defects with high accuracy remains challenging, often leading to false detections. To address these issues, this paper proposes an improved YOLOv5n algorithm for industrial surface defect detection. The main improvements are as follows: the DSConv-CA module in the backbone network enhances the feature extraction capability, the Gold-YOLO structure replaces the original PANet structure in the neck to improve information fusion, and the SIoU loss function is adopted to replace the regression loss, further improving detection accuracy. Experimental results demonstrate that the improved YOLOv5n algorithm achieves a mean average precision of 75.3% on the NEU-DET dataset, which is 4.3% higher than the original model.