Enhanced YOLO for fabric defect detection via region-aware attention and hypergraph computation

Fabric defect detection is vital for textile quality control, yet it remains challenging due to multiscale defect sizes, complex morphologies, and irregular spatial distributions. Existing methods, particularly convolution-based YOLO series, often struggle to effectively capture fine-grained texture variations and long-range dependencies of defects. To address these issues, this paper proposes an enhanced YOLO for fabric defect detection via region-aware attention and hypergraph computation, named FDD-YOLO. Specifically, a Region-Aware Efficient Layer Aggregation Network (RA-ELAN) is proposed to leverage region-aware attention for long-range contextual dependencies modeling, thus enhancing the perception of small and blurry defects under low-contrast conditions without compromising computational efficiency. Additionally, a Mixed Graph Fusion Block (MGF-Block) with a dual-branch structure is introduced to synergistically model global semantic information and fine-grained texture features via hypergraph and lightweight convolutions. Extensive experiments conducted on three datasets demonstrate that the proposed method achieves superior performance and competitive computational efficiency, making it suitable for real-world industrial applications.