Enhancing YOLOv8 for Fabric Defect Detection with Advanced Feature Fusion

Fabric defect detection is critical for textile quality control, but existing deep learning-based methods face significant challenges: detecting small defects, handling complex textures, and maintaining computational efficiency. While recent advances in lightweight convolutions, attention mechanisms, and multi-scale feature fusion have improved detection accuracy, each approach alone has inherent limitations. This paper addresses these challenges by presenting an enhanced YOLOv8-based framework that strategically integrates three complementary components: GhostConvV3 for efficient feature extraction, CA-EDAC2F for optimized C2F with dual-attention fusion, and EnlightTransformerSPPF for adaptive multi-scale fusion and frequency-guided enhancement. Our comprehensive experiments on the Alibaba Fabric Defect Detection dataset demonstrate significant performance gains—improving [email protected] by 6.40% and [email protected]:0.95 by 15.89% over the baseline YOLOv8—while maintaining reasonable computational costs. These results validate our approach’s effectiveness in achieving both accurate and efficient defect detection, making it suitable for real-world textile inspection systems.