基于GSS-YOLOv8n的轻量化织物疵点检测算法  

Lightweight fabric defect detection algorithm based on GSS-YOLOv8n

作　　者：井振威 张团善[1] JING Zhenwei;ZHANG Tuanshan(Xi'an Polytechnic University,Xi'an,710613,China)

机构地区：[1]西安工程大学,陕西西安710613

出　　处：《棉纺织技术》2025年第4期59-66,共8页Cotton Textile Technology

基　　金：国家自然科学基金项目(51735010)。

摘　　要：针对织物疵点检测方式大多为人工操作且检测耗时、背景复杂、所含疵点种类繁多等问题,提出一种改进YOLOv8算法的轻量级检测模型GSS-YOLOv8n。首先,采用了兼顾速度和精度的GSConv替代原有的标准卷积核,并在Neck使用一次性聚合方法来设计跨级部分网络(GSCSP)模块VoVGSCSP替换C2f模块,引入GSlim-Neck结构,降低了计算和网络结构的复杂性,保持了足够的精度;其次,重新设计了YOLOv8检测头SCGD,在降低模型参数量的同时减少了细节特征的遗漏率,提升了检测头定位和分类的性能,提高了模型的鲁棒性。最后,引入损失函数Shape-IoU,该损失函数考虑了边界框回归样本自身的形状和尺寸对边界框回归的影响,使得IoU更加准确和稳健。试验结果表明:GSS-YOLOv8n模型的mAP@0.5和mAP@0.5∶0.95为98.1%、73.5%,相比于原模型分别提高了1.0个百分点和9.7个百分点,参数量和计算量分别降低了34.7%和35.4%,检测速度达到42.4帧/s。GSS-YOLOv8n模型在实现轻量化的基础上可以实时准确地识别织物疵点。Aiming at the problems of fabric defect detection,such as manual operation,time-consuming,complex background and many kinds of defects,a lightweight detection model GSS-YOLOv8n based on improved YOLOv8 algorithm was proposed.Firstly,GSConv was used to replace the original standard convolution core for speed and precision,and one-time aggregation method was used to design the cross-level partial network(GSCSP)module VoVGSCSP to replace the C2f module,the introduction of GSlim-Neck structure reduced the complexity of calculation and network structure,and maintained sufficient precision.Secondly,the YOLOv8 detector SCGD was redesigned to reduce the number of model parameters,while the missing rate of details was reduced.The performance of detection head localization and classification was improved,and the robustness of the model was improved.Finally,the loss function Shape-IoU was introduced,which considered the influence of the shape and size of the bounding box regression sample on the bounding box regression,making the IoU more accurate and robust.The results showed that mAP@0.5 and mAP@0.5∶0.95 of GSS-YOLOv8n model were 98.1%and 73.5%,which were 1.0 percentage points and 9.7 percentage points higher than the original model respectively,the parameters and computation were reduced by 34.7%and 35.4%respectively,and the detection speed reached 42.4 frames/s.GSS-YOLOv8n model could identify fabric defects accurately in real time based on the realization of lightweight.

关 键 词：YOLOv8 织物疵点检测 Shape-IoU 轻量化 GSConv 

分 类 号：TS101[轻工技术与工程—纺织工程]