Computer Vision-Driven Textile Quality Assessment: An EfficientNet Approach for Multi-Class Fabric Defect Recognition
Students & Supervisors
Student Authors
Supervisors
Abstract
This paper presents an automated framework for fabric defect detection using a computer vision pipeline powered by EfficientNet-B0. The system classifies textile samples into six categories: defect-free, hole, horizontal line, vertical line, stain, and other line defects. To enhance robustness under industrial conditions, the dataset of 2,739 images was preprocessed with normalization and augmentation. The proposed model integrates Mobile Inverted Bottleneck (MBConv) and Squeezeand- Excitation (SE) blocks for efficient feature representation and employs weighted sampling to address class imbalance. Experimental results show that EfficientNet-B0 achieves a classification accuracy of 97%, significantly outperforming baseline ANN and CNN models. In addition, Grad-CAM visualizations confirm the interpretability of the framework by highlighting the defective regions attended by the network. The proposed system has strong potential for real-time deployment in the textile sector, offering improved consistency in inspection, reduced operational costs, and seamless integration with production lines of varying scales.
Keywords
Publication Details
- Type of Publication:
- Conference Name: 28th International Conference on Computer and Information Technology (ICCIT 2025)
- Date of Conference: 19/12/2025 - 19/12/2025
- Venue: LONG BEACH HOTEL, COX'S BAZAR, BANGLADESH
- Organizer: IEEE