Zhou, Yizhou (Beijing Jiaotong University), Liu, Ze (Beijing Jiaotong University), Li, Junjie (Beijing Jiaotong University), Cao, Jingming (Beijing Jiaotong University), Ma, Dingyi (Beijing Jiaotong University), Liang, Hao (Beijing Jiaotong University)

Dual-Level Distillation for Rail Defect Detection Based on Electromagnetic Tomography

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on April 2, 2026

Abstract

Electromagnetic tomography (EMT), as a non-contact internal defect detection technology, shows great promise for rail inspection. However, its practical application remains challenged by significant imaging artifacts, difficulty in modeling weak defects, and insufficient coupling between reconstruction and detection tasks. To address these issues, this paper proposes a Dual-Level Distilled Lightweight Detection Framework (DL-DLDF), which combines the structural restoration capability of Attention U-Net and the detection performance of YOLO to achieve cross-task information transfer and collaborative optimization. Specifically, a dynamic channel feature alignment module (DCFA) and a histogram-constrained super-resolution branch (HCSR) are introduced to supervise the reconstruction network from feature and image spaces, respectively, while a zero-cost gradient decoupling strategy ensures optimal computational efficiency at inference. Experimental results demonstrate that DL-DLDF achieves a 10.21% improvement in mAP@0.5 and a 1.7dB gain in PSNR over existing baselines. This study lays a foundation for intelligent, efficient, and scalable defect detection in next-generation railway monitoring systems.