Research on Steel Surface Defect Detection Algorithm Based on Improved YOLOv8n

Abstract

Abstract Steel, as one of the most used materials, is of great importance for steel defect detection in industry. Aiming at the existing existing deep learning-based steel surface defect detection algorithms that have problems such as misdetection, leakage, low detection accuracy and speed, a steel surface defect detection algorithm YOLOv8n-CFP is proposed to improve YOLOv8n. Firstly, the CPCA attention mechanism module is added, which combines the channel attention mechanism with the spatial attention mechanism, to improve the model’s recognition accuracy and generalization ability. After that, the Faster module is used instead of Bottleneck to reduce the computational complexity of the C2f module and lighten the network structure. Finally, the PIoU loss function is used instead of CIoU to optimize the performance of anchor frame regression, which guides the anchor frame to regress efficiently and achieve faster convergence by combining the target size adaptive penalty factor and gradient adjustment function. The experiments show that compared with the basic YOLOv8n network, the YOLOv8n-CFP inspection network improves the mAP from 76.2% to 79.0% on the steel surface defects inspection dataset of YOLOv8n, which is an improvement of 2.8%, and the model volume, parameter count, and computational size are reduced by 17.5%, 18.3%, and 17.3%, respectively, and the FPS value is improved by 60.4. Compared with other YOLOv8n networks, the average accuracy mAP of YOLOv8n-CFP inspection network is increased to 79.0%. Compared with other YOLO algorithms, YOLOv8n-CFP has significant advantages in steel surface defect detection.