Font Size: a A A

Research On Fabric Detection Method Based On Machine Learning

Posted on:2020-02-05Degree:MasterType:Thesis
Country:ChinaCandidate:Y ChenFull Text:PDF
GTID:2381330578959949Subject:Physical Electronics
Abstract/Summary:PDF Full Text Request
Clothing,food,shelter and transportation are necessary for people's daily life.For clothing,fabric is an indispensable raw material for clothing production.Therefore,a large number of fabrics are needed to produce to meet the demand of people.With the improvement of living standards,people's pursuit of fabric quality has become more stringent.Due to improper mechanical operation,objective environment and other factors,fabric defects appear in the production of fabric.Common defects include holes,stains,discoloration,yarn,and so on.One of the key factor to determine the quality of fabric is whether there are defects on the surface of the fabric.The price of faulty cloth is much lower than that of faultless cloth.The decline in prices will lead to lower profits,less income,even to close the company.In order to improve the profits of enterprises,fabric defect detection is needed.Many companies still maintain the manual detection method with naked eye inspection,which will lead to the decline of detection speed,and is unsuitable for actual production.Therefore,it is particularly important to use machine vision to replace the human eye and realize automatic detection of fabric defects.A quickly and effectively detection algorithm to detect fabric defects is the key to realize machine vision detection.With the wide application of machine learning in the fields of image processing and target detection,the realization of the fabric defect detection algorithm becomes possible.This paper proposes a machine-based fabric defect detection algorithm,which aims to better realize the automatic detection of fabric defects.The main research work is as follows:(1)A fabric defect detection algorithm based on feature extraction and support vector machine is proposed.Generally,fabric has the characteristics of periodic variation of surface texture and the uniformity of surface structure and organization,as well as directionality.When faults appear on the fabric surface,the integrity and structure of the fabric texture are destroyed.Compared with the defectfree point,the eigenvalue of the defect point has changed.Here,five visual descriptor algorithms,which include MFS,HOG,SIFT,PHOG and HOG-NMF,are used to describe the eigenvalue of the defect point and defect-free point.Due to the limitation of the fabric production line,the samples of fabric defect image are also limited.So the support vector machine is selected to distinguish the defects,which is suitable for small sample training and can make up for the problem of limited sample size.The algorithm includes feature extraction and classification learning,and has two steps: training and testing.In training,input the fabric block feature and the defect information of the training sample to the support vector machine and obtain the network model.In testing,input the characteristics of the fabric piece to the trained network and judge whether there are defects or not according to the output result of the network.(2)A fabric defect detection algorithm based on convolutional neural network is proposed.The detection method based on feature extraction cannot adapt to learning and has poor versatility since it relies on the feature.The algorithm can improve the validity and efficiency of detection by using the convolutional neural network based on its good self-adaptive learning ability.This paper uses five network models including AlexNet,VGG16,VGG19,GoogleNet and improve CNN to determine whether the fabric contains defects or not.(3)A fabric detection algorithm based on Faster RCNN is proposed.Since the convolutional neural network is suitable for the classification of defects,but can't find the position of the fabric defects the Faster RCNN can be used to find the defect position and distinguish the defect classification.Firstly,the RPN network(Region Proposal Network)is used to determine the candidate window that may contain defects.Then,the classification of the convolutional neural network is used to judge the defect type,and determine the exact position of the candidate window by regression to realize the accurate location and discrimination of defects.The experimental image data in this paper come from TILDA standard fabric image library.Four different fabric types including striped fabric,plaid fabric,twill fabric and plain fabric are selected to verify the algorithm.And four defect type is discoloration stain,yarn,and hole.A large number of experimental results show that the proposed algorithms can effectively detect the fabric defects point out the location and the classification of the defect.And they can provide reference for the enterprise and engineering to realize fabric defect detection.
Keywords/Search Tags:Defect detection, feature extraction, support vector machine, convolutional neural network, Faster RCNN
PDF Full Text Request
Related items