Research On Fabric Defect Detection Algorithm

The appearance of fabric is composed of colors,patterns and texture,and various types and different forms of defects lead to fabric defect detection ineffectively.Therefore,an innovative,efficient and high-speed fabric defect algorithm based on machine learning is urgent need to develop to improve the accuracy of the automatic fabric defect detection and achieve the objective evaluation of fabric defect level.Thence the studying on fabric defect detection algorithm is important to the current textile printing and dyeing industry.This paper focuses on the research on the yarn-dyed fabric,patterned fabric and grey fabric defect detection algorithms,the main research contents are as follows:In order to detect the defects of yarn-dyed fabric and patterned fabric,an algorithm based on the dual-dictionary is presented,which is learned by convolutional matching pursuit and K-singular value decomposition based on Gabor.First of all,mean sampling as preprocessing is conducted for the sampled images to reduce the impact of fabric texture background information on the fabric defect detection process.After that,the optimal sliding window is used to select defect-free image blocks as the input sample set of convolutional matching pursuit,so the dual-dictionary can be obtained by the wavelet fusion of the convolution dictionary and the small-scale dictionary,and sparse coding is applied to train the dual-dictionary.At last,the method takes the defect-free and defective fabric image's projections in the dual-dictionary as features,respectively,and the Euclidean distance are calculated to gain the fabric defect detection results.In view of the problem of yarn-dyed fabric and patterned fabric defect detection,a method based on alternating direction method with Gaussian back substitution(ADMG)image decomposition is employed.Firstly,histogram equalization as preprocessing is conducted for the sampled images to reduce the image noise and improve the signal-to-noise ratio.Secondly,ADMG image decomposition method based on the combination of the total variation norm and semi-norm in negative Sobolev space is employed,and the Yarn-dyed fabric images could be decomposed into defect structure u and texture structure v.Finally,the adaptive fuzzy threshold method is applied to segment the defect part u to identify the defect area.A new detection approach is proposed to detect various uniform and structured grey fabric defects based on the multiple Gabor filters and kernel principal component analysis.First of all,images are filtered by multiple Gabor filters with 6 scales and 4 orientations to extract texture features.Secondly,the sub-blocks divided from the texture features are spliced and the high-dimension data reduced by kernel principal component analysis.At last,the similarity matrix is calculated by Euclidean norm and binarized with threshold method.