Rail Surface Defect Detection with Multi-Scale Detection Heads and Receptive Field Enhancement

Rail surface defects have diverse types and large scale variations, and standard detectors often miss fine-grained texture defects such as corrugation. This paper proposes a lightweight rail defect detection model based on an improved YOLOv11n. Three progressive improvements are introduced. First, a high-resolution P2 detection head is added to the neck to build a four-scale detection architecture, enhancing the perception of fine-grained textures. Second, C3k2 modules in the deep backbone are replaced with C2f modules, which refine gradient flow and deep semantic feature extraction while alleviating false detections caused by the shallow head. Third, an SPPF module is introduced in the P2 branch to enlarge the receptive field and compensate for weak semantics in shallow features. On a self-built Rail Surface Defect Dataset, the final model achieves 0.797 [email protected] and 0.562 [email protected] on the independent test set, improving the baseline by 11.78% and 14.93%, respectively. For corrugation, which is severely missed by the baseline, accuracy increases from 0.292 to 0.581. The model size increases only from 5.51 MB to 6.07 MB, achieving a good balance between accuracy and deployment cost for high-frequency, large-scale automated railway inspection.

Authors

Institutions

Publication Details

Journal
Academic Journal of Emerging Technologies
Published
2026-11-18
DOI
https://doi.org/10.63313/ajet.9081
Primary Topic
Infrastructure Maintenance and Monitoring
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Rail Surface Defect Detection with Multi-Scale Detection Heads and Receptive Field Enhancement

Xu Jie, Zhichao Hu, Jincheng Li, Taoyu Wang et al.
2 links
Academic Journal of Emerging Technologies
Infrastructure Maintenance and Monitoring
article

Rail Surface Defect Detection with Multi-Scale Detection Heads and Receptive Field Enhancement

Xu Jie, Zhichao Hu, Jincheng Li, Taoyu Wang, Yuhao Ma
article en

Abstract

Rail surface defects have diverse types and large scale variations, and standard detectors often miss fine-grained texture defects such as corrugation. This paper proposes a lightweight rail defect detection model based on an improved YOLOv11n. Three progressive improvements are introduced. First, a high-resolution P2 detection head is added to the neck to build a four-scale detection architecture, enhancing the perception of fine-grained textures. Second, C3k2 modules in the deep backbone are replaced with C2f modules, which refine gradient flow and deep semantic feature extraction while alleviating false detections caused by the shallow head. Third, an SPPF module is introduced in the P2 branch to enlarge the receptive field and compensate for weak semantics in shallow features. On a self-built Rail Surface Defect Dataset, the final model achieves 0.797 [email protected] and 0.562 [email protected] on the independent test set, improving the baseline by 11.78% and 14.93%, respectively. For corrugation, which is severely missed by the baseline, accuracy increases from 0.292 to 0.581. The model size increases only from 5.51 MB to 6.07 MB, achieving a good balance between accuracy and deployment cost for high-frequency, large-scale automated railway inspection.

Academic Journal of Emerging TechnologiesVol. 4(1)
Tianjin University of Technology and Education (CN)
Industry, innovation and infrastructure
Infrastructure Maintenance and Monitoring
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon