基于SAC-YOLO的轻量化织物疵点分割算法

打开文本图片集

中图分类号：TS101.97;TP391.41 文献标志码：ADOI：10.13338/j. issn. 1674-649x.2025.03.008

Lightweight fabric defect segmentation algorithm based on SAC-YOLO

ZHANG Zhouqiang ，LI Cheng ，WANG Kangxu，CHEN Furong ，CUI Fangbin （Schoolof Mechanical and Electrical Engineering，Xi'an Polytechnic University，Xi'an 71O048，China）

Abstract Aimed at the problems of low detection accuracy and complex model in the process of fabric defect detection，a lightweight fabric defect segmentation algorithm based on SAC-YOLO was proposed. Firstly，a lightweight down-sampling module，ADown，with stronger extraction capabilities，was introduced in the Backbone part to improve accuracy while reducing model parameters.SENetv2 attention mechanism was then incorporated into the C2f module to enhance the model's ability to extract semantic and detailed information of small defect targets. Secondly，the Neck layer uses a multi-scale feature fusion module，significantly reducing the model parameters and computational load. Finally，the Inner-CIoU loss function was introduced to replace the original CIoU bounding box regression loss function，accelerating the model convergence speed and improving the model generalization ability. Experimental results show that compared to the improved YOLOv8s，the proposed model reduces the parameter count by 45.8% and the computational load by 8.8 billion floating-point operations. The mAP50 and mAP50-95 of defect detection are improved by 0.5% and 1.7% ，respectively. The precision，recall， mAP50 ，and mAP50-95 of segmentation are improved by 1.3%，0.2%，0.4% ，and 1.4% ，respectively. Compared with other mainstream segmentation algorithms，the improved algorithm model shows superior performance in both detection and segmentation.

Keywords fabric defect detection; YOLOv8； image segmentation; attention mechanism

0引言

纺织品瑕疵检测是确保纺织产品质量的关键环节，随着技术的发展，自动化检测逐渐成为研究的热点[1-3]。（剩余15453字）