Research on Lightweight Surface Defect Detection Network Model Based on Improved Yolov8

The deep learning-based surface defect detection method exhibits advanced detection capability. Aiming at the problems of large networks, high computing power and difficult deployment on edge devices, this paper presents a method for detecting surface defects using YOLOv8n. By substituting the original YOLOv8n backbone with lightweight FasterNet module, redundant information processing is reduced and model speed is improved. Use lightweight cross-scale feature fusion modules to enhance the fusion of low-level and high-level features, further reducing the amount of computation and parameters while maintaining accuracy. The multi-head self-attention mechanism has been incorporated to enhance the model's effectiveness. Test results indicate that, compared to YOLOv8n, the improved method achieves the same mean average precision (mAP) of 76.6% on the NEU-DET steel surface defect dataset. However, the number of model parameters and the computational load have been decreased by 69.9% and 44.8% respectively, with the model size now being only 28.6% of the original. These findings demonstrate that the improved model is more lightweight and better suited for use on hardware with constrained computing resources.