Fabric Defects Detection And Classification Technology Based On Wavelet And Statistical Learning Theory

As a means of quality assurance, Fabric defect detection is mainly done by hand,work intensity and low efficiency, it’s a good choice to choose automated machineinstead. In this paper, we investigate two aspects: fabric defect detection andclassification. The main work is shown as following:Singular value decomposition (SVD) is an effective algebraic feature extractionmethod. Firstly, this paper studies the defect detection and classification algorithmsbased on SVD. In the defect detection phase, we make full use of the informationgenerated by singular value decomposition, including the singular values and singularvalue feature vectors, to achieve defect segmentation. Then we extract features based onsingular value and apply BP neural network classifier to achieve defect classification.For the irregular shape and texture features of the defects, we study the defectdetection and classification algorithms based on Gabor wavelet network (GWN) andsupport vector machine (SVM). We are approaching the background texture of thefabric image, and construct the optimal Gabor filter to achieve defect segmentation.Then we extract defects features based on GLCM, and apply SVM to achieve defectclassification. It achieved good detection and classification results.For some minor flaws, we separately investigate the defect detection algorithmbased on stationary wavelet transform (SWT) and non-sampling direction filteringbanks (NSDFB). It completes independently multi-scale decomposition andmulti-directional decomposition in two steps. It allows to choose the number ofdirections decomposition in different scales and provides a flexible multi-directionaland multi-scale expansion.