A Reliability Aware Graph Mamba Framework for Intelligent Sensor-Based Landslide Detection and Prevention
Year: 2026
Type: conference-paper
Abstract: In disaster management, landslide detection enables timely identification of vulnerable regions and unstable slopes before catastrophic failures. Accurate and continuous detection mechanisms play a crucial role in reducing human casualties, infrastructure damage, and economic losses. Existing landslide detection approaches failed to handle multimodal environmental information and lack the ability to capture long-term spatio-temporal risk evolution. Additionally, these approaches failed to adapt to heterogeneous monitoring conditions. To overcome these challenges, a Reliability Aware Mamba Network (ReMambaNet) is proposed for landslide detection and prevention. Initially, RGB images and environmental monitoring variables are collected. In preprocessing stage, image resizing, min max normalization, linear interpolation and z score normalization is applied. After that, reliability aware weighting and cross modal attention fusion generates feature representations. The fused representations are transformed into dynamic graph which are passed through the graph mamba to capture complex spatio-temporal dependencies and long-range interactions among monitoring regions. Finally, an adaptive risk memory bank stores and retrieves historical high-risk patterns for risk aware representation learning in landslide detection and risk-aware assessment. Experimental demonstrates that the proposed ReMambaNet achieved an f1-score of 96.02% and 97.07% on Niangniangba and Hokkaido datasets, respectively.
Language: English
Subfield: Management, Monitoring, Policy and Law
Field: Environmental Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7203473401