Design of 60 Steel Rail Surface Damage Recognition and Detection System Based on Deep Learning

View at publisher Summarise with VIRAT
Authors: Yu Wang, Hetian Mo, Mengting Ying, Jianxing Liu, Haoxing Chen, Zihao Chen, Jiongfeng Liang, Jiang-Tao Hu, Bingren Peng
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This study focuses on the demand for surface damage detection of 60 steel rails under the background of expanding railway transportation scale. To address the problems of low efficiency and poor accuracy of traditional manual detection, a deep learning based intelligent recognition and detection system is designed and implemented. Firstly, steel rail surface images are collected under multiple working conditions using an industrial line array camera. The data quality is improved through preprocessing such as grayscale conversion, filtering and denoising, histogram equalization, and normalization, Furthermore, a deep learning model with an improved ResNet-34 as the backbone network is constructed, incorporating deformable convolution to enhance local feature capture capability. The FPN multi-scale feature fusion structure and dual branch output architecture (synchronously implementing damage type and location prediction) are combined to optimize the loss function through weighted cross entropy and Smooth L1 joint optimization. Actual railway field tests have shown that the system has high detection accuracy and low false alarm rate, significantly better than traditional manual features or shallow machine learning methods. It provides reliable technical support for railway track maintenance and has important application value in ensuring operational safety and reducing maintenance costs.

Fields of Study & Subjects

Infrastructure Maintenance and Monitoring Industrial Vision Systems and Defect Detection Railway Engineering and Dynamics

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
564–569
Publication date
28 January 2026
Language
EN