Loading article...
Loading article...
Stream/Area:
-
Subject:
-
Paper ID:
IJSREM62208
Volume:
10 |
4 |
April 2026
Pages:
1-7
Publication Frequency:
Monthly (12 issues per year)
License:International Journal of Scientific Research in Engineering and Management follows the Creative Commons Attribution 4.0 (CC BY 4.0) License, allowing free use and distribution with proper attribution.
Artificial Intelligence And Data Science
Dr. Akhilesh Das Gupta Institute of Professional Studies
Artificial Intelligence And Data Science
Dr. Akhilesh Das Gupta Institute of Professional Studies
Artificial Intelligence And Data Science
Dr. Akhilesh Das Gupta Institute of Professional Studies
The accurate detection of heterogeneous terrain features, such as landslides , remains a significant challenge in remote sensing due to complex topographical variations and extreme class imbalance. In this paper, we propose a novel multi-scale deep learning framework powered by a customized 9-channel topo-spectral feature stack. This stack fuses globally normalized Sentinel-2 multispectral optical bands (Red, Green, Blue) and vegetation/moisture indices (NDVI,NDMI) with Digital Elevation Model (DEM) data (Slope, Elevation). Crucially, we explicitly model terrain-vegetation dynamics through engineered interconnection features (Slope-NDVI and Slope-NDMI) to capture physical precursors of terrain failure.We feed this spatial data into an optimized Attention ResUNet architecture enhanced with Atrous Spatial Pyramid Pooling (ASPP), enabling the simultaneous capture of localized pixel signatures and wide-area contextual topography. To combat spatial overfitting and enforce context-invariant learning, we implement a simple data augmentation strategy adapted for multi-channel physical data, optimized via a hybrid BCE-Dice loss function. Empirical evaluations demonstrate that our proposed model achieves a strong balance of precision and recall, yielding a peak F1 score of 0.750 and an Intersection over Union (IoU) of 0.597. Furthermore, post-inference probability threshold calibration reveals optimal boundary extraction at a 0.40 confidence threshold, significantly recovering false negatives along ambiguous topographical edges. This research demonstrates that integrating multi-scale spatial pooling with engineered topo-spectral interaction features provides a highly robust solution for automated terrain segmentation.
KEYWORDS
We have presented a deep learning approach for landslide segmentation from remote sensing imagery. By evaluating the effect of various improvements of feature engineering, network architecture, loss function and post processing. We finally construct the ASPP-ARUNet, which bases on Attention Residual U-net architecture. The extensive experiments prove that our proposed ASPP-ARUNet is robust on LandSlide4Sense dataset, which shows potential to apply remote sensing image based landslide analysis system.
[1] Wikipedia, "https://en.wikipedia.org/wiki/Landslide" [2] 5 killed, 11 missing as fresh cloudbursts, landslides hit uttarakhand,"https://www.ndtv.com/india-news/5-killed-11-missing-as-fresh-cloudbursts-landslides-hit-uttarakhand-9184621". [3] Introduction to Landslide Evaluation Tools- Mapping, Remote Sensing, and Monitoring of Landslides, "https://pubs.usgs.gov/circ/1325/pdf/Sections/AppendixB.pdf" [4] Predicting Landslide Using Machine Learning Techniques, "https://www.researchgate.net/publication/382302854_Predicting_Landslide_Using_Machine_Learning_Techniques" [5] Hesheng Chen, Yi He, Lifeng Zhang, Wang Yang, Yaoxiang Liu, Binghai Gao, "A Multi-Input Channel U-Net Landslide Detection Method Fusing SAR Multisource Remote Sensing Data ", December 05-2032 [6] Yingxu Song, Yujia Zou, Yuan Li, Yueshun He, Weicheng Wu, Ruiqing Niu and ShuaiXu, "Enhancing Landslide Detection with SBConv-Optimized U-Net Architecture Based on Multisource Remote Sensing Data", June 12-2024 [7] Landslide4Sense: Multi-sensor landslide detection competition & benchmark dataset, "https://github.com/iarai/Landslide4Sense-2022" [8] Omid Ghorbanzadeh, Yonghao Xu, Hengwei Zhao, Junjue Wang, Yanfei Zhong, Dong Zhao, Qi Zang, Shuang Wang, Fahong Zhang, Yilei Shi, Xiao Xiang Zhu, Lin Bai, Weile Li, Weihang Peng, and Pedram Ghamisi, “The outcome of the 2022 landslide4sense competition: Advanced landslide detection from multisource satellite imagery,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 9927–9942, 2022[9] Rekha R.Nair, Tina Babu, Tripti Singh, Afnaan K, "Unet and Semantic Segmentation Based Landslide Detection System," Internatinal Conference on Computing for Sustainable Global Development (INDIACom), August 21-2025 [10] Syed Mujtaba Hussaine, Linlong Mu, Yimin Lu & Syed Sajid Hussain, "Landslide Image Segmentation with Attention Residual U-Net: A Hybrid Deep Learning Model",Procedia Computer Science, vol. 258, pp. 2029-2039, 2025
Publication Date
5/1/2026
Peer Review Type
Double-Blind Peer Review
Review Rounds: 2
Documents
Article Views
0
Total views since publication
PDF Downloads
3
Total downloads since publication
Geographic Reach
5
Digital Preservation: This article is permanently archived on Zenodo and complies with all data preservation standards across major indexing platforms.
Plagiarism Screening
This manuscript has been checked for plagiarism using TurnitinPlagiarismChecker software.
Conflict of Interest
The authors declare no conflict of interest related to this research work.
Funding Information
No funding information available for this research.
Related research in the same subject area
Loading similar articles...