Surface Defect Detection Algorithm for Strip Alloy Material Based on Improved YOLOv8

Abstract

In response to the problems of missed detection, false detection, and slow detection speed in surface defect detection of strip alloy material, this paper proposes an improved YOLOv8 based algorithm for surface defect detection of strip alloy material. To enhance multi-scale feature fusion, we first propose the Multiscale Feature Encoder (MFE) module. Additionally, we construct the MultiScale Feature Aggregation and Diffusion (MFAD) structure in the model’s neck, which effectively disseminates rich contextual features across multiple scales. Moreover, the Task Dynamic Alignment Detection Head (TDADH) is introduced in the model’s head, leveraging parameter sharing and task alignment mechanisms to reduce model complexity while improving detection accuracy. Finally, we incorporate the Perceptual Attention Spatial Pyramid Pooling (PASPP) module at the end of the backbone, utilizing an explicit dynamic selection mechanism through attention mechanisms to enhance the model’s feature representation capability. Experimental results demonstrate that our approach achieves a detection accuracy of 90.1% mAP50, with a parameter count of 2.543M and a detection speed of 232 fps, outperforming existing methods on our dataset. Furthermore, our model also exhibits superior performance on the GC10-DET and PASCAL VOC2012 datasets. Our codes and pretrained models are available at https://github.com/yw-xx/Improved-YOLOv8.