Entropy-driven two-stage flash-and-flag cascade for rail defect inspection with yolo-based detection and GPS tagging on standard laptops

Name: Karin Vitoonkijvanit

Organization : TAIST-Science Tokyo. Dept. of AIoT

Email : karin.vit@ku.th

Name: Sanparith Marukatat

Organization : National Science and Technology Development Agency. Image Processing and Understanding Team

Email : sanparith.marukatat@nectec.or.th

Name: Kanjanapan Sukvichai

Organization : Kasetsart University. Dept. of Electrical Engineering

Email : fengkpsc@ku.ac.th

Name: Isshiki, Tsuyoshi

Organization : Institute of Science Tokyo. Dept. of Information and Communications Engineering

Email : isshiki@ict.e.titech.ac.jp

LCSH: Railroad tracks -- Inspection

LCSH: Computer vision

LCSH: Deep learning (Machine learning)

LCSH: Global Positioning System

Abstract: Railway track defects-ranging from hairline cracks and faulty joints to surface corrosion-continue to jeopardize safety and drive up maintenance costs. In response, we developed Tetsumi Suite, a custom GUI-based inspection toolkit that implements a straightforward two-stage pipeline running live on a standard laptop (Intel i5-11300H, 16 GB RAM, NVMe SSD), with optional CUDA acceleration (GeForce MX450) for batch processing. In Stage I, a lightweight, CPU-resident filter rapidly scans multi-camera frames by downsampling and applying fast heuristic checks-histogram equalization, gradient magnitude, edge density, texture entropy, and optional background subtraction-to flag only the most suspicious images without impeding operations. Flagged frames are then forwarded to Stage II, where a TorchScript-optimized YOLO11m model (CPU or GPU) assigns detailed labels (Crack, RCF, Bad Joint, Fixed, or Normal) and overlays precise bounding boxes. All saved frames are automatically embedded with GPS EXIF metadata, enabling seamless geospatial visualization. Evaluated on a reserved set of 200 hand-annotated images, Tetsumi Suite achieves mAP@0.5 = 0.989 and F₁ = 0.950 at a confidence threshold of 0.567. To the best of our knowledge, this represents the first end-to-end solution that integrates real-time heuristic filtering, deep-learning-based classification, and GPS tagging on a standard laptop-delivering a scalable, high-accuracy tool suitable for field-ready rail inspection."

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2025

Modified: 2025-12-26

Issued: 2025-12-26

บทความ/Article

application/pdf

BibliograpyCitation : In IEEE Thailand Section. Antennas and Propagation/Electron Devices/Microwave Theory and Techniques Joint Chapter, Electrical Engineering/Electronics, Computer, Telecommunications, and Information Technology Association, and King Mongkut's University of Technology North Bangkok. The 5th Research, Invention and Innovation Congress (RI2C 2025) (pp.1-6). Bangkok : IEEE Thailand Section

eng

©copyrights King Mongkut's University of Technology North Bangkok

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	RI2C 2025pp.1-6.pdf	2.31 MB

ใช้เวลา

0.033314 วินาที