Kumaşlarda dokuma hatası tespiti icin spektral alanda kümeleme tabanli gürbüz bir yöntem

The purpose of this dissertation is to design an automated system using machine vision techniques for inspection of faults in textile industry. Detection of defects in fabrics has a substantial importance to prevent delivering faulted products. Because of human factor drawbacks like fatigue, boredom and oversight, manual inspection falling behind expectations of the industry. There are countless automated systems in the literature, however, most of them rely on machine learning methods where defected and defect free images are fed to the system for training or the approaches have numerous parameters to be tuned. For these reasons supervised systems are difficult and discomfort to use. In this study, Fourier Transform based unsupervised approach used to inspect a fabric. The method does not require any prior knowledge about the fabric pattern or the defect class. Proposed algorithm obtains the spectral representation of the partitioned image and measures the distance between each part and a reference representation obtained from the same inspected sample. Although the literature crowded with automated inspection systems, a suitable dataset for textile fabric could not be found. Existent databases either not freely available or has lack of defect classes. Therefore, a dataset contains 26 defect class plus a clean class collected. The database called Eskisehir Technical University Textile Defect Dataset (ESTD) and consist of 2969 samples of ten diverse type of fabric.