Automated Fabric Defect Detection and Localization Using YOLOv8 and Deep Learning Techniques

Fabrics as fiber-based materials, play a crucial role in the production of different types of clothing. However, large-scale production processes often lead to the occurrence of defects, making quality control a significant challenge. This study presents a deep learning-based approach for fabric defect detection using the YOLOV8 model, alongside other established deep learning models like CNN, VGG16, InceptionV3, and MobileNetV2. Three extensive datasets are used for experimentation and evaluation, a custom dataset comprising mobile camera images of different types of fabrics and two public datasets obtained from Kaggle and GitHub. Deep learning models were applied to the two datasets, custom, and Kaggle to extract their features. InceptionV3 outperformed the other models, acquiring a test accuracy of 97% on the primary and 98% on the secondary dataset, while the modified custom CNN model acquired a test accuracy of 94% on the primary dataset. Furthermore, YOLOV8 applied to the GitHub dataset, demonstrated superior capabilities in localizing and identifying defective regions on fabrics. This research highlights the effectiveness of DL models in enhancing defect detection precision and paves the way for more efficient quality control in textile manufacturing.