Improved YOLOv5s Fabric Defects Based On Deep Learning Detection Method Research

Textile intelligent manufacturing is the development direction of today’s textile industry,and fabric defect detection is an important part of fabric quality assurance.At present,most of the fabric defect detection commonly used by enterprises adopts manual detection,but manual detection has low detection efficiency and high labor intensity,and cannot meet the needs of real-time online detection.With the rapid development of machine vision,some enterprises have applied machine vision to the textile field and developed a machine vision system related to fabric defect detection,which has effectively improved the defects of manual detection.However,most of the existing fabric defect detection systems use statistical methods,structural methods and other fabric defect detection algorithms,which have shortcomings such as low detection rate and poor self-adaptation,and cannot meet the requirements of real-time detection.In recent years,deep learning algorithms have made a lot of progress in the field of image processing,with strong applicability.Therefore,the development of an improved YOLOv5 s fabric defect detection system based on deep learning in this paper has important industrial application value and significance.The main research work and results of this paper are as follows:(1)Explain the research background and significance of fabric defect detection,analyze the domestic and foreign product technology status and fabric defect detection algorithm,compare the traditional algorithm with the deep learning algorithm,discuss the advantages and disadvantages of fabric defect detection,and propose the depth Learning applied to fabric defect detection.(2)Analyze the fabric texture theory and common defect types,and analyze the causes of small target fabric defects.The advantages and disadvantages of common target detection algorithms are analyzed.Among them,YOLOv5 has a faster detection speed while ensuring a certain accuracy.Therefore,this paper finally determines YOLOv5 as the basic algorithm.Taking the fabric defects collected by a textile factory in Zhejiang as the research object,the data is expanded by using affine transformation,gray value adjustment and noise transformation,and a fabric defect data set is established.(3)Using the YOLOv5 s algorithm in the deep learning target detection algorithm as the basic network,the algorithm is improved for the problems of high missed detection rate of small target fabric defects,low defect location accuracy,and high network loss rate.The specific method is to add a GAM attention mechanism module between the 8th and 9th layers in the backbone network,and add a GAM attention mechanism module after the 17 th,20th,and 23 rd layers of the neck network;the GIOU The loss function is replaced by the SIOU loss function;the coupling head of the original detection end is changed to a decoupling detection head,and the classification task and the regression task are separated;the structure of YOLOv5 s is modified,and a small target detection layer is added to reduce the need for small target fabrics.Defect miss rate.The research results show that the YOLOv5s-GSD4 L algorithm proposed in this paper improves the mean average precision m AP by 7.1%compared with the original YOLOv5 s algorithm,reaching 97.6%.At the same time,in the detection of actual fabric defects,the missed detection rate of small targets is low,and the detection accuracy is higher.(4)Based on the YOLOv5s-GSD4 L algorithm fabric defect detection algorithm proposed in this paper,the fabric defect visual detection software and hardware platform solution are developed,and the functions are integrated on one interface,which can facilitate users to detect fabric defects and view defect detection results.(5)Summarize the full text and put forward the prospect for the next research direction.