Automatic inspection system for defects classification of stretch knitted fabrics

Fabric defect detection and classification plays a very important role for the automatic detection in fabrics. This study refers to the four common seen defects of stretch knitted fabrics: laddering, end-out, hole, and oil spot. First of all, wavelet transfer is applied to obtain its wavelet energy to take them as defect features of this image, and then the back-propagation neural network (BPNN) was used to carry out the defects classification of the fabrics. In addition, by using the Taguchi method combined with BPNN had improved the deficiency of BPNN, which requires overly time consuming trial-and-error to find the learning parameters, and therefore could converge even faster, having an even smaller convergence error and better recognition rate. Experimental results have proven the final root-mean-square error convergence of the Taguchi-based BPNN was 0.000199, and the recognition rate can reach 96.5%.