Home // GEOProcessing 2026, The Eighteenth International Conference on Advanced Geographic Information Systems, Applications, and Services // View article

Railway Sinkhole Detection: An Integrated Approach Combining Geomorphological Segmentation and Descriptor-Based Machine Learning

Authors:
Maryem Bouali
Fakhreddine Ababsa
Rani El Meouche
Muhammad Ali Sammuneh
Bahar Salavati
Flavien Viguier

Keywords: LiDAR Point Clouds; Digital Elevation Models (DEM); Geomorphological Analysis; Supervised Learning.

Abstract:
Railway sinkholes represent a critical geotechnical hazard for rail infrastructure safety and require reliable detection methods based on high-resolution topographic data. This paper proposes an integrated methodology combining geomorphological analysis and machine learning for the automated detection of railway sinkholes from Light Detection and Ranging (LiDAR)- derived Digital Elevation Models (DEMs). The proposed frame-work relies on a two-stage approach. First, a geomorphological processing chain is applied to generate high-resolution DEMs and delineate topographic depressions using watershed segmentation. This step enables the systematic extraction of candidate basins potentially corresponding to sinkholes. Second, each extracted basin is characterized using a set of handcrafted geometric and morphometric descriptors capturing properties, such as shape, depth, volume, and spatial structure. These features are then used within a supervised learning framework to discriminate true sinkholes from other terrain irregularities. Several classification algorithms, are implemented and compared to assess their performance and robustness. Experimental results demonstrate that integrating feature-based machine learning with geomorphological segmentation significantly improves detection accuracy and reduces false positives compared to a purely morphological approach. The proposed workflow contributes to spatial terrain analysis and infrastructure monitoring by providing a scalable and data-driven methodology for railway risk assessment based solely on elevation data.

Pages: 6 to 12

Copyright: Copyright (c) IARIA, 2026

Publication date: May 24, 2026

Published in: conference

ISSN: 2308-393X

ISBN: 978-1-68558-384-2

Location: Venice, Italy

Dates: from May 24, 2026 to May 28, 2026