Research on Lightweight Fabric Defect Detection Algorithm Based on Improved YOLOv8

To overcome the limitations of low accuracy, excessive parameters, and high computational load in YOLOv8-based fabric defect detection, this work introduced an enhanced algorithm built upon an improved YOLOv8 framework. Our approach introduced key structural improvements to the YOLOv8 baseline model through targeted algorithmic adjustments. SEAMHead was adopted within the detection module, replacing the original head to enable hierarchical feature analysis for fabric defect detection. For the feature fusion layer, the HS-FPN module was introduced to replace the original feature fusion module, effectively reduced information attenuation of crucial features while suppressing noise interference and substantially enhanced the network's resistance to texture interference and directional adaptability. Significant performance improvements were observed in the proposed algorithm versus YOLOv8: a 0.4 percentage point enhancement in and a 0.8 percentage point increase in :0.95, while reducing the parameter size by 42% and FLOPs by 29%. In comparison with mainstream object detection algorithms and advanced fabric defect detection algorithms, the proposed improved algorithm exhibited superior performance across comprehensive metrics, enabling efficient fabric defect detection.