Pilling analysis for textile grading of synthetic fibre material using image processing and machine learning techniques

Abstract

Textile pilling is a persistent issue that results in an unattractive surface on garments, impacting their aesthetic and commercial value. This study employs image processing and machine learning techniques to grade fleece textiles based on pilling evaluation. Two approaches were explored: the first utilized a discrete Fourier transform combined with Gaussian filtering, while the second employed Daubechies wavelets. Binarization was used to isolate textile pilling from the surrounding fabric area. In this study, morphological and topological image processing techniques were applied to extract key features from the image data, creating a comprehensive database for fabric analysis. Machine learning techniques, specifically Support Vector Machine (SVM) and Artificial Neural Networks (ANN), were then used to objectively address the textile grading problem. The Fourier-Gaussian approach achieved classification accuracies of 95.67% with ANN and 92.34% with SVM, while the Daubechies wavelet approach yielded accuracies of 94.23% and 90.67%, respectively. In terms of pilling detection, the Fourier-Gaussian method identified 67 instances of pilling with a pilling area of 1674 units, whereas the Daubechies wavelets method detected 56 instances with a pilling area of 654 units. This automated system enhances textile quality assessment and production efficiency by effectively detecting and quantifying pilling.