The effect of wavelet transform for fabric defect classification

Abstract

An automatic control system during fabric production improves production quality. For this reason, the number of automatic systems developed is increasing day by day. These systems use different methods, different data sets, and different approaches. We investigate the effects of wavelet transform using four different feature sets (wavelet-based Principal Component Analysis, wavelet-based Gray Level Co-occurrence Matrix, Principal Component Analysis and Gray Level Co-occurrence Matrix). The methods of K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) are used as classifiers. Experiments have been carried out for six different fabric defects (fly, dirty warp, tight-loose warp, fibrous weft, rub mark and weft stack) on 57 images. The experimental results performed show that wavelet-based PCA (Principal Component Analysis) and KNN have been optimal methods for achieving the highest success rate. We have achieved a 92.9825% accuracy rate by using them.