Identification of twill grey fabric defects using DC suppressed Fourier power spectrum sum features

Defect identification and classification has been a focal point in fabric inspection research, and remains challenging because of new microstructure defects occurring in twill grey panting fabrics weaved on modern looms such as Air jet looms and Rapier looms. The twill fabric defects that occur commonly on these auto looms are mostly localized microstructure defects such as looseweft and stitches. This paper focusses on the application of DC suppressed Fourier power spectrum obtained from Fourier Transform for the analysis of fabric images in terms of significant frequency contents, which depict the periodicity of fabric along with their magnitudes, magnitude sums between peaks and the fabric cover factor of the woven fabric, in order to identify the fabric faults. The analysis was carried out on real twill weave grey fabric of different fabric specifications by collecting as many as 27 statistical features along with fabric cover factor obtained from the marginals of DC suppressed Fourier power spectrum which were used as inputs to the neural network implementing Levenberg-Marquardt Back-propagation algorithm. The results of the neural network, optimized with 27(40) neurons in the input, a hidden layer and 3(2) neurons in the output layer respectively for the two fabric classes namely S1(S2), for identification of grey fabric defects are encouraging. The neural network converged in less than 35 iterations and gave a classification accuracy of almost 100% when compared to the NN classification rate of 89.28% without considering fabric cover factor. The details of the experimentation and the results thereof are presented in this paper.