基于改进YOLOv7的织物疵点检测算法  

Weaving fabric defect detection algorithm based on improved YOLOv7

作　　者：毋涛[1] 崔青 殷强 邓魏永 梁芷 WU Tao;CUI Qing;YIN Qiang;DENG Weiyong;LIANG Zhi(School of Computer Science,Xi’an Polytechnic University,Xi’an 710048,China;China National Textile And Apparel Council,Beijing 100020,China;Huayu Zhengying Group,Jinjiang 362200,Fujian,China)

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048 [2]中国纺织工业联合会,北京100020 [3]华宇铮蓥集团,福建晋江362200

出　　处：《纺织高校基础科学学报》2023年第4期29-36,共8页Basic Sciences Journal of Textile Universities

基　　金：国家自然科学基金面上项目(62176204)。

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YOLOv7算法的轻量级检测方法。首先,在主干和颈部引入FasterNet结构,在保证检测精度的同时又降低网络参数量;其次,为减少位置信息丢失,在特征提取阶段引入CA注意力模块,以提高网络的表达能力;最后,引入新的损失函数Focal-EIoU,将Focal与EIoU相结合,提高疵点的分类和定位精度。通过对构建的含有6种疵点的面料数据集进行测试可以看出,相比于原算法,所提算法计算量GFLOPS降低至38.6,参数量降低6.14×10^(6),平均精度均值提高4.6%,漏检率降低5.5%,帧率达到63.2帧/s。A lightweight detection method based on improved YOLOv7 algorithm was proposed to address the issues of manual operation,time-consuming,complex background,and a wide variety of defects in fabric defect detection methods.Firstly,a FasterNet structure was introduced into the backbone and neck to ensure detection accuracy while reducing the amount of network parameters.Secondly,in order to reduce the loss of location information,CA mechanism was introduced in the feature extraction stage to improve the expressive capacity of the network.Finally,a new loss function Focal-EIoU was introduced,and combined with EIoU to improve the classification and positioning accuracy of defects.By testing the fabric dataset constructed in this paper containing six types of defects,it can be seen that compared to the original algorithm,the proposed algorithm reduces the computational complexity of GFLOPS to 38.6,reduces the number of parameters by 6.14×10^(6),increases the mAP value by 4.6%,reduces the miss detection rate by 5.5%,and achieves a frame rate of 63.2 f/s.

关 键 词：YOLOv7 织物疵点检测 FasterNet 注意力机制 Focal-EIoU 

分 类 号：TS101.9[轻工技术与工程—纺织工程]