Fabric defect detection system based on image processing for circular knitting machines

Fabric, having an indispensable use in daily life, is produced in weaving and knitting machines. Various disadvantages which occur during manufacture lead to the defects in the produced fabric. As a result of this, losses in raw materials, labour and energy occur in the textile industry. Defect detection systems that prevent the production of defected fabric are available in weaving machines. However, the absence of a product to detect the fabric defect in circular knitting machine (with the desired success) is the main motivation of this thesis. In this thesis, a fabric defect detection system that can operate in real time on a circular knitting machine has been developed. This system comprises (1) the establishment of the image acquisition device; (2) the construction of a fabric database; (3) development of defect detection methods; (4) defect detection during the production process. As a result of the studies conducted in the thesis process, a new fabric database that contains 6 different fabrics and 10 different fabric defects has been built. The features of these images in the spatial and frequency domain have been obtained, and classification of these images has been carried out. The most important contribution of this thesis is that it gives the literature 6 new feature extraction methods: GDF-HOG, Eig(Hess)-HOG, second order HOG, Eig(Hess)-CoHOG, GM-CoHOG and surface labeling-based CoHOG. The advantages and shortcomings of each method in comparison with conventional methods are discussed in detail in the thesis. In addition to these methods in the spatial domain, by using Fourier, wavelet and shearlet transform methods, specific statistical features of the fabric images have been extracted and their classification has been provided. Artificial Neural Networks have been used as a classifier. The improved defect detection system has been set on the circular knitting machine in Madoksan Textile Ltd. Com. (Malatya). In real time defect detection studies, common fabric defect types have been detected successfully during production.