Fabric Defect Detection Algorithm Based On Texture Feature And Low Rank Representation

Due to the complexity of fabric surface defect patterns and texture features,automatic detection of fabric defects based on machine vision is a challenging task,which has been widely concerned by relevant researchers.In recent years,low rank representation model(LRR)has been proved to be suitable for fabric defect detection,which can represent fabric image as sparse object and redundant background respectively.However,the detection performance of the method based on low rank representation is limited by the effective representation of the fabric image and the effectiveness of the model.Therefore,this thesis studies the texture representation method and low rank model,and proposes a series of related fabric defect detection algorithms.1)The fabric defect detection algorithm based on handcrafted feature and nonlocal total variation regularization term low rank decomposition is proposed.Firstly,the fabric image to be tested is divided into blocks,and the handcrafted features are extracted from each block,after that,all the block features are integrated into a feature matrix representing the whole fabric image.Then,the nonlocal total variation regularization term is introduced to construct the low rank decomposition model,and the alternating direction multiplier method is used to solve the model.Finally,according to the sparse matrix obtained by the solution,the defect saliency map is generated through the spatial correspondence,and the defect region is located after segmentation.The robustness and effectiveness of the proposed algorithm are proved by experiments.2)The fabric defect detection algorithm based on shallow network and non-convex low rank representation is proposed.Firstly,a shallow convolutional neural network(SNET)with three convolution layers is established to represent the fabric image.Then,a non-convex low rank representation(NLRR)model is constructed to effectively separate the defects from the background.In addition,double low rank matrix representation(DLRMR)is used to fuse saliency maps generated by different convolution layer features.Experimental results show that the proposed method has high computational efficiency and detection accuracy,and double low rank fusion further improves detection performance.3)The fabric defect detection method based on deep-handcrafted feature and weighted low rank matrix representation is proposed.By fusing the global deep feature extracted from VGG16 network and the handcrafted low-level feature,the feature representation ability is effectively improved.Then,the weighted low rank matrix representation(WLRR)model is constructed,which gives different weights to the singular values of different feature matrices and shrinks them.The most significant features of fabric texture are retained,the defects are highlighted effectively and the background is suppressed.Finally,the salient image is segmented by threshold segmentation algorithm and the defect segmentation image is obtained.Experimental results show that the proposed method has high adaptability and detection accuracy,and is superior to the existing methods.