Survey of fabric defect detection using UNet architecture

Abstract

Due to the rapid growth of new computer technologies, fabric defect identification has become a prominent topic in recent years. Detection of defects in the fabric is a vital stage in the textiles industry’s quality control. Traditional fabric inspections frequently use manual visual processes leading to inaccurate and imprecise results that are unsuitable for long-term industrial use. Many researchers have contributed their efforts for fabric flaw detection systems using different techniques such as computer vision, image processing, etc. All the previous work done in this field has some limitations. The accuracy rate in the existing work ranges from 88% to 90%. Datasets used by many systems had fewer images and were typical of one defect type. Also, pre-processing techniques were absent in the previously proposed system. This research presents a modified U-shaped network (U- Net), an enhanced convolutional neural network for detecting fabric defects. The attention mechanism is established based on network size compression. In this methodology, the U-Net network is improvised to discover the fabric defect more accurately and precisely.