SYSTEMATIC REVIEW OF DEEP LEARNING TECHNIQUES FOR REAL-TIME DEFECT DETECTION IN FABRIC MATERIALS

Abstract

This systematic review delves into the significant advancements of deep learning techniques in realtime defect detection for fabric materials, a critical component of improving quality control in textile manufacturing. By providing an in-depth analysis of key methods, including Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), Autoencoders (AEs), and Active Continual Learning (ACL) frameworks, the review highlights how these technologies have dramatically enhanced the accuracy, efficiency, and adaptability of defect detection processes. The integration of these techniques has led to more precise identification of defects, reducing the reliance on manual inspection and significantly improving production speeds and quality. However, the review also identifies several persistent challenges, such as the need for more realistic and diverse synthetic data to better train models, the complexities in maintaining model adaptability to evolving and unforeseen defect types, and the difficulties associated with seamlessly integrating these sophisticated systems into existing industrial workflows without disrupting current operations. The study underscores the importance of ongoing research to refine deep learning methods, aiming to enhance their robustness, scalability, and reliability across various industrial environments, from small-scale textile workshops to large-scale manufacturing plants. This review not only synthesizes the current progress in deep learning applications for fabric defect detection but also outlines critical areas for future exploration, offering a detailed roadmap for the continued evolution of automated quality control in the textile industry, ultimately leading to more consistent and higher-quality textile products