<p>Rail defect detection is essential for ensuring railway safety and enabling reliable maintenance. However, existing methods often struggle in complex environments, where feature ambiguity and occlusion significantly degrade detection performance, leading to frequent missed and false detections. To overcome these limitations, we propose a topology-enhanced relational modeling framework that jointly captures local, global, and high-order dependencies. Specifically, a semantic association graph is constructed to enable dynamic multi-order feature extraction, allowing the model to learn structured relationships among spatial regions and effectively enhance the representation of subtle defects. To further improve global perception, a multi-directional state-space modeling module is introduced to capture long-range dependencies and enhance spatial sensitivity. Moreover, a hypergraph-based interaction mechanism is designed to model complex high-order relationships, where hypergraph convolution facilitates deep feature fusion and improves discrimination of heterogeneous defect patterns. Experimental results demonstrate that the proposed method consistently outperforms state-of-the-art approaches, achieving higher accuracy and stronger robustness under challenging conditions.</p>