Optimized algorithm for steel surface defect detection based on YOLOv8n-CSDE

Surface defect detection remains a significant challenge in the industrial sector due to frequent inspection errors, false positives, and low detection accuracy. Traditional object detection algorithms often face issues such as missed detections, false detections, insufficient feature extraction capabilities, and inadequate feature fusion when applied to steel surface defect detection tasks. To address these challenges, this study proposes an improved surface defect detection method for industrial components, named YOLOv8n-CSDE, based on the YOLOv8n object detection framework. Firstly, the CGAttention is incorporated into the C2f module to enhance the model's semantic understanding and feature generation capabilities. Secondly, the DySample method is employed to replace the nearest-neighbor algorithm, reducing computational complexity and improving up sampling efficiency. Thirdly, the original Spatial Pyramid Pooling Fast module is upgraded to the SimSPPF module to enable the network to better emphasize the target's features. Finally, the CIoU loss function in YOLOv8n has been replaced with the EIoU loss function to accelerate model convergence and improve localization performance. Extensive experiments conducted on the NEU-DET dataset demonstrate that the enhanced model achieves an accuracy of 76.6%, representing an improvement of 2.8% over traditional models. The experimental results indicate that the proposed method achieves more accurate detection of steel surface defects compared to state-of-the-art object detection algorithms.