Distinguish Some Textile Defects by Adopting Image Processing Methods and Intelligence Techniques

Abstract

The automatic control of the fabric is one of the important steps in the spinning and weaving industry in order to preserve the quality of the fabric. The manual methods have been used for decades to control the product using human vision. The monitoring process is very strenuous, time consuming and cost effective. To reduce the costs required there arise the needing of automated systems appearance to examine, detect and apply tissue defects. The aim of the proposed work is to build an efficient system for detecting and classifying textile defects using advanced image processing techniques based on new methods of combining the practical implementation of image segmentation and features extraction, as well as the use of artificial intelligence techniques of neural networks for detection and classification. The system was built in two phases: the first is the defect detection phase, and the second phase is the classification phase, where live images were collected as a textile database from the textile factory in Mosul as well as the local market. The fabrics were carefully selected and these fabrics are of different types and colors, some of these have no defect at all and some of them have up to fourteen types of defects. 560 images were collected; 280 of which were non defective fabrics, 280 were defective, and there are 20 images for every type of defect, at the defect detection phase, the statistical second-class attributes of the GLCM matrix (energy, variance, correlation, homogeneity) are extracted, while in the classification phase, the statistical first-class attributes, mean and skewness, and the geometric attribute of the total defect size. Two neural networks were used as determinants of detection and classification: the Back Propagation Neural Network (BPNN) and the Elman network. The proposed system showed a 95.7% discrimination rate compared with other similar work in the same field.