Improved YOLOv8-based surface defect detection algorithm for aluminum profile

Aiming at the issues of small surface defects on aluminum profiles and low detection accuracy, this paper proposes an improved aluminum profile surface defect detection method based on YOLOv8. Firstly, the Ghostv2 module is introduced to minimize the number of parameters and volume of the model, and the convolutional layers of YOLOv8 are replaced with the Spatial Depthwise-separable Convolution module (SPD-Conv) to facilitate feature fusion and cross-scale information interaction in the Convolutional Neural Network (CNN), thus improving the ability to extract small target features. Secondly, a detection module for small targets is integrated into the Head section to improve small target detection effectiveness. Then, the LskBlock attention block from LSKNet is introduced into the neck network to dynamically adjust the spatial receptive field to better simulate the range context of different objects. Ultimately, the CIoU loss function is substituted with Wise-IoU, which has stronger positioning ability for detection targets compared to the original CIoU, improving the accuracy of target detection. Experiments show that the improved network model achieves an mAP@0.5 of 84.3%, representing a 5.8% improvement compared to the original YOLOv8 algorithm.