Treffer: An integrated texture and depth isomorphic imaging and cross‐modal network for rail surface defect detection and measurement.

Rail defects significantly impact train operations, even posing serious safety risks. Existing methods can automatically collect images from the rail surface and identify apparent defects while facing challenges such as high false positive rates, visually subtle defects omit errors, and quantitative defect size measurement. To address these issues, an integrated 2D&3D rail surface defect detection and measurement framework is proposed. Initially, this framework introduces an isomorphic imaging system with a long–short exposure mechanism, which uses a single camera to capture pixel‐level registered 2D texture and 3D depth images in a single imaging procedure. Subsequently, a cross‐modal defect detection network is developed to explore complementary semantic and structural information from 2D and 3D images hierarchically, enhancing defect identification capability. Finally, considering the physical curvature changes of the railhead, a partition projection‐based 3D measurement method is established to provide accurate quantitative measurements for defect depth, width, and length. This study collects 2045 operational rail surface images with visible defects and establishes a standard 2D&3D defect detection dataset to validate model performance. Experimental results show that this technology achieves improvements of 7.26% and 9.17% in maximum F1‐score and recall, compared to prevalent SAINet. The defect depth measurement accuracy reached 0.18 mm. Extensive experiments on publicly available non‐service rail surface defect datasets also demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]