High-performance multimodal approach for defect identification in knitted and woven fabric

Abstract

Fabric inspection is a key quality assurance process in the garment industry as it involves the detection of defects in a fabric roll prior to being sent for production. Many studies have been conducted on defect identification in either knitted or woven fabrics, but only a few have considered both types. In this paper, a method for detecting defects in both knitted and woven fabrics is proposed. The method involves extracting co-occurrence, wavelet and local entropy features from a fabric image and classifying the image as defective or defect-free using a classifier with these features given as input. Five commonly-used classifiers were tested. This method was applied to a dataset with seventeen different types of defects and an overall classification accuracy of 93.31% was achieved by the k-nearest neighbours classifier.