Automated Quality Control in Textiles: A Computer Vision Blended Approach to Fabric Defect Detection

This research proposes a novel deep learning architecture, R-ECA-GeM-Net, for automated fabric defect detection by integrating Residual Efficient Channel Attention (R-ECA) blocks, BlurPool-based downsampling, and Generalized Mean (GeM) pooling to enhance feature extraction and preserve fine-grained texture information. The model was evaluated on a combined dataset of 9,855 fabric images across six classes and achieved a test accuracy of 97.67% with a macro F1-score of 0.9770, demonstrating its effectiveness for textile quality control. Additionally, the study incorporates Grad-CAM and LIME to provide visual explanations of model predictions, improving the transparency and interpretability of deep learning-based defect inspection systems.