Fabric defect detection of industrial knitting machines using YOLOv7

Abstract

The textile industry, one of the largest industries of Bangladesh, relies heavily on the quality of yarn for the production of fabrics. Like other products, the yarn is subject to defects that declines the quality of fabrics. The traditional yarn defect detection technique is done through manual inspection, which is time-consuming, laborious, and prone to human error. The application of automation processes in this sector can accelerate quality control to a huge extent. This paper deals with the use of image processing using YOLOv7 in the field of yarn defect detection. For this task, we've created a custom dataset with three types of defects. Using this custom dataset, our method had a mean average precision of 87%. Furthermore, we have implemented the model in real-time using a camera device, and it was able to detect defects with a highest confidence rate of 86%. Computer vision has already been used in healthcare and transportation, but its use in the garment industry is rare. This paper explores the implementation of a machine learning algorithm to find defects on the yarn surfaces to increase the quality of fabrics and decrease the production of defects.