Fabric defect classification using wavelet frames and minimum classification error-based neural network

Abstract

This paper presents a new method for fabric defect classification by using a wavelet frames feature extractor and a minimum classification error-based neural network. Channel variances at the outputs of the wavelet frame decomposition are extracted to characterize each non-overlapping window of the fabric image, which is further assigned to a defect category with a neural network classifier. In our work, a Minimum Classification Error (MCE) criterion is used in the training of the neural network for the improvement of classification performance. The developed defect classification method has been evaluated on the classification of 329 defect samples from nine types of defects and 82 non-defect samples, where an 93.4% classification accuracy was achieved.