RFA-YOLOv8: Steel Plate Surface Defect Detection Algorithm Based On Improved YOLOv8

Abstract To address the issues in existing steel plate surface defect detection algorithms, such as low accuracy, numerous model parameters, and high computational complexity, this paper introduces a novel algorithm named RFA-YOLOv8, which is built on an enhanced version of YOLOv8n. The approach enhances the C2f module within the backbone network and incorporates a receptive field attention (RFA) mechanism to refine the convolution layer, leading to the development of the C2f_RFAConv structure. This advancement significantly boosts feature extraction capabilities in complex scenarios. Experimental results demonstrate that RFA-YOLOv8 achieves a 74.9% accuracy on the NEU-DET dataset, surpassing YOLOv8n by 1.8%, with minimal changes in model parameters and floating-point operations. Moreover, improvements in both recall and precision rates confirm that RFA-YOLOv8 delivers high detection accuracy and efficiency, making it valuable for industrial quality inspections.