Intelligent Railway Block System Design with Lightweight CNN and RFID

• Published in: CACS International Automatic Control Conference (Online) pp. 1 - 6
• Main Authors: Chang, You-Yi, Hung, Kuei-Jung, Perng, Jau-Woei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 31.10.2024
• Subjects: Artificial intelligence; automatic train tracking; block and signal control; Bluetooth information systems; Computational complexity; Computational modeling; Convolutional neural networks; feature extraction; Lightweight CNN; Monitoring; Radiofrequency identification; Rail transportation; railway obstacle detection; Real-time systems; RFID technology; Safety; System analysis and design
• ISSN: 2473-7259
• DOI: 10.1109/CACS63404.2024.10773268

Obstacles on railways present significant safety risks. In particular, Taiwan Railways (TR) often faces natural and man-made obstructions. While advanced artificial intelligence (AI) models offer precise detection, their high computational demands limit deployment. Thus, lightweight convolutional neural network (CNN) models have been developed, balancing detection accuracy with hardware efficiency. Utilizing feature-extracting preprocessing, these models achieve F 2 scores above 0.89 and recalls over 0.9, while maintaining computational complexity under 15 MFLOPs and model sizes within 25 kB. This ensures compatibility with standard PCs and low-cost, low-power systems, enabling widespread deployment in regions lacking advanced AI monitoring. Additionally, TR's aging block and train-tracking systems, which are prone to failures, are enhanced with a power-saving radio frequency identification(RFID) system compatible with the CNN detector and Bluetooth-enabled passenger displays. This modernizes TR's infrastructure, improving safety operations and travel-information reliability.