Improved Yolov8 Algorithm for Detecting Industrial Defects on Aluminum Sheet

Aiming at the problems existing in the target detection of defects on the surface of aluminum sheet, including: small target defect detection is prone to leakage, the imbalance between model detection speed and accuracy, etc., an efficient and accurate detection algorithm, MBW-YOLOv8n, is proposed. on the basis of YOLOv8 algorithm, a lightweight Multidimensional Collaborative Attention is introduced between the neck and the detection head. Based on the YOLOv8 algorithm, a lightweight Multidimensional Collaborative Attention is introduced between the neck and the detection head to enhance the localization differentiation between small and dense targets; the C2f module in the detection head is replaced by a bitemporal feature aggregation module, which aggregates a variety of sensory field features and improves the algorithm's anti-jamming property; and finally, the Normalized Wasserstein Finally, the Normalized Wasserstein Distance metric is embedded into the original loss function to replace the commonly used IoU metric, which further improves the performance of YOLOv8 small target detection and enhances the robustness to different scale targets. Experiments show that the average accuracy of MBW-YOLOv8 is improved by 1.7% compared with the original YOLOv8 model, which verifies the robustness of the algorithm in detecting industrial defects on the surface of aluminum sheet in industrial production.