BLW-YOLOv8s: a model for detecting steel defects in industrial scenarios

To improve detection accuracy for indistinct features and small targets in steel surface defect detection, a YOLOv8sbased algorithm is proposed. To address blurred features caused by background interference in complex environments, the SPPF_LSKA module, inspired by Large Separable Kernel Attention, is introduced to enhance feature representation. To reduce missed detections of small defect targets, Bi-directional Feature Pyramid Network (BiFPN) is utilized to redesign the bottleneck, enhancing small target detection while decreasing parameters and computational complexity by 24.46% and 10%, respectively. Additionally, the WIoUv3 loss function, with its dynamic non-monotonic focusing, replaces CIoU to improve localization accuracy in complex conditions. Experimental results show that the improved model increases [email protected] by 3.5% on the NEU-DET dataset, with significant reductions in parameters and complexity, making it suitable for industrial applications.