Surface-defect segmentation in latent space
Reduced computation by segmenting defects in the decoder latent space, with improved few-shot accuracy over the evaluated baselines.
IEEE Transactions on Instrumentation and Measurement, 2024 ↗Computer vision · Railway infrastructure
Mingxu Li
Assistant ResearcherChina Academy of Railway Sciences
Research Institute
Computer vision for railway inspection
I study how to detect small defects and segment their boundaries with limited data and computing resources. My work spans railway catenary and rail-surface inspection, from low-power screening to detailed recognition.
Latest publication · IEEE TITS, 2026
S2R: from screening
to recognition
Low-power screening triggers detailed segmentation when a potential defect is found. S2R combines spiking and convolutional neural networks, with contrastive learning and hard-sample mining to distinguish defects from background interference.
Read the paper
Selected research results
Efficient, detailed perception
RBNet for railway defect segmentation
A rendering-inspired architecture that moves from a coarse prediction to fine defect boundaries while keeping the model lightweight.
IEEE Transactions on Instrumentation and Measurement, 2023 ↗Background
About me
I received my Ph.D. in Computer Science from Southwest Jiaotong University in June 2026. During my doctorate, I completed joint research at Nanyang Technological University (August 2024–December 2025) and the University of Birmingham (January–August 2024).
I also hold an M.Eng. in Software Engineering and a B.Eng. in Internet of Things Engineering. My other research includes road-crack recognition, multi-vehicle tracking and contour-guided restoration of Thangka artworks and Tibetan murals.
Education, projects and honors ↗Research & industry experience
China Academy of Railway Sciences
Assistant Researcher · Railway visual inspection
The Alan Turing Institute
AI Research Intern · Participated in data analysis for a Mastercard prediction project
Tencent WeChat Group
Test Development Intern · Automated JSAPI testing for WeChat Work
Research papers
Selected publications
-
2026
S2R: From Screening to Recognition, Event-Driven Perception of Railway Defects in Low-Power Systems
IEEE Transactions on Intelligent Transportation Systems
-
2024
Latent Space Segmentation Model for Visual Surface Defect Inspection
IEEE Transactions on Instrumentation and Measurement · vol. 73, pp. 1–11, Art. no. 5029111
-
2023
RBNet: An Ultrafast Rendering-Based Architecture for Railway Defect Segmentation
IEEE Transactions on Instrumentation and Measurement · vol. 72, pp. 1–8, Art. no. 2512808
-
2023
BVTracker: Multivehicle Tracking Based on Behavioral-Visual Features
IEEE Sensors Journal · vol. 23, no. 11, pp. 11815–11824
Contact
For research collaboration and professional enquiries.