基于改进YOLOv7的织物表面疵点检测技术  

Fabric Surface Defect Detection Technology Based on Improved YOLOv7

作　　者：任经琦 张团善[1] 赵浩铭 REN Jingqi;ZHANG Tuanshan;ZHAO Haoming(School of Mechanical and Electrical Engineering,Xi'an Polytechnic University,Xi'an 710600,China)

机构地区：[1]西安工程大学机电工程学院,陕西西安710600

出　　处：《沈阳大学学报（自然科学版）》2024年第2期112-120,F0003,共10页Journal of Shenyang University：Natural Science

基　　金：国家自然科学基金资助项目(51735010);西安现代智能纺织设备重点实验室项目(2019220614SYS021CG043)。

摘　　要：针对目前纺织工业中织物疵点检测技术的局限性,提出一种用于自动检测织物缺陷的改进YOLOv7算法。首先,在颈部网络引入SPD-Conv模块,在进行卷积下采样时保留与疵点相关的特征辨别信息,解决了原网络对于小目标的特征信息学习不足的问题;其次,YOLOv7的主干网络通过引入CA注意力机制,在兼顾通道注意力的同时,还考虑了位置信息,从而更有效地识别疵点;最后,把WIoU用作边框损失函数,使其更加侧重于一般品质的锚框,从而增强了YOLOv7的泛化能力。通过实验对比发现,改进后算法的mAP值为92.28%,精度为95.65%,分别比原始YOLOv7算法提高了2.64%和4.12%,能够达到纺织产业在疵点检测方面的要求。In view of the limitations of the current fabric defect detection technology in the textile industry,an improved YOLOv7 algorithm for automatic detection of fabric defects was proposed.Firstly,the SPD-Conv module was introduced into the neck network,which retained the feature discrimination information related to defects during convolution downsampling,and solved the problem of insufficient learning of feature information of small targets in the original network.Secondly,the backbone network of YOLOv7 introduced the CA attention mechanism,which not only took into account the channel attention,but also considered the location information,so as to identify defects more effectively.Finally,WIoU was used as the border loss function to make it more focused on the anchor box of general quality,so as to enhance the generalization ability of YOLOv7.Through experimental comparison,it was found that the mAP value and accuracy of the improved algorithm were 92.28%and 95.65%,which were 2.64%and 4.12%higher than the original YOLOv7 algorithm,respectively,which could meet the requirements of the textile industry in terms of defect detection.

关 键 词：疵点检测 YOLOv7 SPD-Conv模块 WIoU CA注意力机制 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论] TS107[自动化与计算机技术—计算机科学与技术]