Abstract:
Detecting fabric defects is a crucial step in guaranteeing the quality and pricing of
textiles. Defects in fabric result in a considerable amount being discarded as waste,
leading to substantial annual losses. While manual inspection has traditionally been
the norm, the adoption of automatic defect detection schemes based on deep learning
models offers a timely and efficient solution for addressing production-related issues
and assessing fabric quality. Extensive research has been conducted on fabric defect
detection methods using deep learning to enhance both production efficiency and product quality. In real-time manufacturing scenarios, datasets lack high quality, precisely
positioned images. Moreover, both plain and printed fabrics are being manufactured in
industries at the same time. So, training a robust model that detects defects in fabric
datasets generated during production with high accuracy using recent deep learning
technologies is required. In this study a real-time dataset sourced directly from Chenab
Textiles, providing authentic and diverse images representative of actual manufacturing conditions is used to train a YOLOv8 model. For comparison, a YOLOv5 and
MobileNetV2-SSD FPN-Lite model is also trained on the same dataset. YOLOv8 produced best results with an mAp of 84.8% on 7 different defect classes for printed and
plain fabrics combined.
