An Extensive Analysis of Textile Defect Detection Using Deep Learning for Automated Quality Inspection

Abstract

The rapid advancement of intelligent textile manufacturing has sped up the development of automatic fabric fault detection systems. The present textile industry is increasingly moving toward intelligent textile production, and the development of autonomous fabric flaw detection systems is speeding up. This can assist lower production costs and inspection times while also improving product quality. With the recent development of Deep Learning (DL), features representation, defect localization and classification accuracy have been improved greatly, making automated quality inspection more reliable than the traditional manual inspection. In this review, the authors present a thorough review of representative textile defect detection techniques, such as the FabricNET microscopic image dataset, Faster Region-Based Convolutional Neural Network (Faster R-CNN), You Only Look Once Version 4 (YOLOv4), and Vision Transformer (ViT). Furthermore, some of the most popular preprocessing and feature extraction methods are analyzed to bring them to the fore for their role in the identification of defects for the purposes of a robust system. The reviewed approaches are comparatively analyzed based on the detection methodology, computational efficiency, capability of feature learning, advantages and disadvantages. The analysis results show that the transformer-based architectures are best suited to the high-resolution image analysis of textiles, while the one-stage detection models are best suited to real-time inspections. The research directions identified and future challenges outlined highlight the need for lightweight, interpretable and scalable intelligent inspection systems for the next generation textile quality control.