Deep Learning Based Fabric Defect Detection

Abstract

Ensuring quality standards is a crucial stage within the textile sector. Automated classification of the fabric defects is a vital step during the fabric manufacturing process in order to prevent any faulted fabric from being supplied to the market. The defects on the surface of the fabric were manually identified by the individuals but this poses problems in terms of human-error and is also time-consuming. Efforts have been made to achieve better precision in defect detection through image processing studies, leading to the development of automated systems. In this study, some high-performing deep learning models are applied including ResNet and VGG-16 and illustrated how these algorithms can be used in the domain of textile manufacturing for fabric defect detection. A combination of images are used ranging from patterned and textured to plain for better defects recognition on any given fabric. The algorithm VGG-16 has displayed 73.91% accuracy while the ResNet algorithm has shown 67.59% accuracy.