Zero-Shot Texture Analysis and Regression-Based Deformation Recognition for Rail Anomaly Detection

This paper presents a novel anomaly detection framework for rail systems, integrating zero-shot texture analysis and regression-based deformation recognition to monitor rail defects effectively. Unlike traditional methods requiring extensive labeled datasets, our zero-shot learning approach operates without pre-labeled examples, making it particularly suitable for rail applications where diverse defect examples are scarce and costly to acquire. We introduce a dual strategy, employing texture anomaly detection for surface defects and regression analysis for geometric deformations, enhancing both the scope and accuracy of anomaly detection. Our methods leverage high-resolution imaging and advanced computational techniques to automate rail integrity assessments continuously. The effectiveness of the proposed framework is rigorously validated through extensive tests using newly developed datasets that encompass a wide range of anomaly scenarios. The results demonstrate significant improvements in early detection of potential rail defects, achieving a 95.67% accuracy in surface anomaly detection and a 94% accuracy in geometry anomaly detection. These findings highlight the robustness and practical applicability of our approach in enhancing rail safety and maintenance operations.