Stereoscopic Saliency Detection for Rail Surface Defects via Global Low‐Rank Reconstruction and Depth Outlier Fusion
Electronics Letters · 2026-01 · 62 (1)
DOI: 10.1049/ell2.70623 ↗Kurz gefasst
Fachartikel in der Zeitschrift Electronics Letters, online erschienen am 20.05.2026. Verfasst von Zhiwen Xiong und 1 weiteren Autor:innen. Die Arbeit nennt 7 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Rail surface defect detection is critical for ensuring transportation safety, yet conventional vision‐based methods struggle with low‐texture defects and complex backgrounds. In this letter, we propose an unsupervised stereoscopic saliency detection method that fuses global low‐rank reconstruction with depth outlier analysis. A binocular line‐scanning system is employed to simultaneously acquire high‐resolution RGB images and dense depth maps. For 2D images, we introduce a Global Low‐Rank Non‐negative Reconstruction (GLRNNR) algorithm incorporating background priors and non‐negative constraints to enhance saliency computation. For depth data, a ruled‐surface fitting model is developed to detect geometric outliers based on rail profile characteristics. The two modalities are collaboratively fused by integrating depth saliency as an additional feature dimension in an iterative refinement process. Experiments on our self‐constructed dataset of 300 stereoscopic image pairs demonstrate that the proposed method achieves an AUC of 0.94 and an MAE of 0.09, outperforming 15 state‐of‐the‐art algorithms. The results validate the effectiveness of multimodal fusion for robust defect detection in industrial scenarios.
Zitieren
APA
Xiong, Z., & Li, Y. (2026). Stereoscopic Saliency Detection for Rail Surface Defects via Global Low‐Rank Reconstruction and Depth Outlier Fusion. Electronics Letters, 62(1). https://doi.org/10.1049/ell2.70623
Vancouver
Xiong Z, Li Y. Stereoscopic Saliency Detection for Rail Surface Defects via Global Low‐Rank Reconstruction and Depth Outlier Fusion. Electronics Letters. 2026;62(1). doi:10.1049/ell2.70623
BibTeX
@article{Xiong2026,
author = {Xiong, Zhiwen and Li, Yuanchun},
title = {Stereoscopic Saliency Detection for Rail Surface Defects via Global Low‐Rank Reconstruction and Depth Outlier Fusion},
journal = {Electronics Letters},
year = {2026},
volume = {62},
number = {1},
doi = {10.1049/ell2.70623}
}
EUVIMED