Aluminum Surface Defect Detection Method Based on Improved YOLOv8

In order to solve the problems of low detection accuracy, poor perception ability of small targets and large calculation parameters in the detection of aluminum surface defects, an improved algorithm based on YOLOv8 was proposed, YOLO-STK. Firstly, combined with the star operation and LSKA (Local Self-Attention) attention mechanism, the C2f-STK module was proposed to enhance the detection ability of aluminum surface defects in complex environments and improve the feature learning performance of the algorithm model. Secondly, a new lightweight backbone network STKNet is designed based on the C2f-STK module, which effectively reduces the granularity loss of target features and reduces the complexity of the model. Finally, in order to make up for the shortcomings of the model in small target detection, an LDB target detection head was proposed to enhance the semantic expression ability of the small target feature layer. The experimental results show that compared with the original YOLOv8 model, the missed detection and false detection of the improved YOLOv8-STK model are significantly improved, the average accuracy is increased by 9.6%, and the number of parameters and calculations are reduced by 26.7% and 22.2%, respectively.