Landslide Hazard Prediction using Satellite Imagery and DL Approaches

Abstract

Artificial intelligence (AI) techniques for landslide prediction in hilly areas, where natural and man-made factors can cause significant damage. While machine learning and feature-based classification methods have been applied to satellite imagery for landslide detection, achieving fully automatic and accurate predictions remains challenging due to data limitations and model performance. This study reviews 50 papers on machine and deep learning algorithms, highlighting research gaps and proposing a novel approach using a modified ResNet101 deep learning model. The proposed model achieves 100% accuracy on an augmented Beijing dataset of more than 4000 satellite images. The findings offer insights into the latest techniques, research gaps, and potentialadvancements in landslide classification using satellite images, providing a resource for researchers and encouraging innovation in this domain.

Introduction

Landslides are serious geological hazards causing loss of life, property damage, and environmental disruption globally. Triggered by factors such as heavy rainfall, earthquakes, volcanic activity, and human actions, landslides pose challenges in timely detection and monitoring. Traditional methods like manual surveys, aerial photography, and remote sensing are useful but often slow, labor-intensive, and limited in coverage.

Recent studies have applied various machine learning (ML) techniques, including Extended Local Binary Patterns (ELBP), Seeded Region Growing (SRG), Support Vector Machines (SVM), and Bayesian learning, mainly for classifying satellite and hyperspectral images. However, fully automatic landslide detection remains difficult due to insufficient quality training data, impacting classification accuracy.

The proposed system employs a deep learning model based on ResNet101, enabling fully automatic and highly accurate landslide detection from satellite imagery. This approach improves on previous systems by leveraging advanced feature extraction and achieving 100% accuracy on an augmented dataset, reducing the need for manual input.

Implementation involves collecting diverse data sources such as satellite images, topographic and geological maps, rainfall records, and sensor data. Deep learning, especially Convolutional Neural Networks (CNNs), automatically learns complex features, enhancing detection capabilities. The system can generate real-time alerts and risk maps to support disaster management.

Key modules used include TensorFlow, Keras, NumPy, OpenCV, and optionally Scikit-learn. Algorithms incorporate CNNs, transfer learning with ResNet101, data augmentation, and the Adam optimizer. The workflow involves data preprocessing, model training, evaluation, and deployment for automated landslide identification.

Conclusion

This project effectively predicts the accuracy of the landslide prediction using satellite imagery and deep learning approaches with accuracy 100%. And the integration of AI techniques into landslide prediction using imagery satellite offers a promising path towards increased resilence to natural disasters.it also useful for thenatural disaster management.And implement systems for continuouslymonitoring the model’s performance in real-world performance and retraining it with new data as it becomes available in satellite imagery and evaluating the model on more diverse and independent datasets.

References

[1] Malviya and Gupta Used Extended Local Binary Patterns (ELBP)and SVM for classifying 24 satellite image classes, addressing noise and unique properties in satellite images. [2] Byun et al. Applied a Seeded Region Growing (SRG) method for landcover classification using multispectral images with efficient segmentation and noise removal techniques. [3] Huang and Zhang Proposed a multi-feature SVM model combining spatial and spectral features for classification. [4] Mianji et al. Introduced a feature reduction method combined with Bayesian learning for hyperspectral image classification.

Copyright

Copyright © 2025 Mrs. V. Ratnasri, G. Dhanalakshmi, P. Devi Prasanna, U. Alekhya, T. Bavana. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.