International Research Journal of Engineering and Technology (IRJET) Volume: 12 Issue: 11 | Nov 2025
www.irjet.net
e-ISSN: 2395-0056 p-ISSN: 2395-0072
Artificial Intelligence Techniques for Landslides Prediction Using Satellite Imagery Prof. Shobha S. Biradar 1, Chitralekha2 1, Professor, Master of Computer Application, VTU’s CPGS, Kalaburagi, Karnataka, India 2, Student, Master of Computer Application, VTU’s CPGS, Kalaburagi, Karnataka, India
------------------------------------------------------------------------------***---------------------------------------------------------------------------------reliable predictions of landslide prone areas using deep Abstract-The research project is developed on the basis of learning technologies will be beneficial. using High-Res. Satellite Imagery and Deep Learning technology to predict landslides. It applies Image Processing 3. OBJECTIVES techniques to enhance the quality of satellite imagery and implements Fine-Tuned ResNet101 deep learning architecture to classify images into Landslide vs NonThe main goal of the project is to create an automated and Landslide categories. The system is also deployed online as accurate system for predicting landslides using deep an app for users (and) Administrators to utilize. The learning and satellite imagery. This includes preexperimental results show that using the Deep Learning processing high-resolution satellite images to improve technology offers a 96.88% accuracy in predicting landslide data quality and using a customized version of the events which demonstrates the validity of the proposed ResNet101 model for classifying images as either landslide method for Disaster Risk Management & Early Warning or no landslide. Additionally, this project also includes Systems. developing a web-based system that enables users to upload images and receive timely predictions. This project also aims to develop a prediction model with high 1. INTRODUCTION accuracy, enabling early warning systems, assisting governmental agencies in effective disaster risk Landslides occur as a result of serious natural disasters in management/mitigation. mountain and hilly areas, particularly in the Himalayas. Increased construction activity and natural disasters, such 4. RESEARCH METHODOLOGY as heavy rain or earthquake activity, will increase the potential for slope failure. The traditional methods to monitor susceptible slopes are slow to develop and can be The research methodology utilizes a structured approach inefficient if limited in scope. In the past few years, to build an automated landslide forecasting model through satellite technology and the application of Artificial the use of deep learning techniques and satellite imagery. Intelligence (AI) have made it possible to provide accurate The first stage of this methodology is acquiring a collection and timely predictions/forecasts of landslides at a much of high-quality satellite images obtained from reputable larger scale than before. This project takes advantage of sources and enhancing the model's ability to generalize by these advancements by using a fine-tuned ResNet101 augmenting these images (inverse transform). The images model and image pre-processing tools along with Deep undergo various preprocessing methods to remove noise, Learning to enhance the quality of detection of landslides normalize them, resize them and segment them into and provide supporting elements of early warning smaller sections. The processed images are then used to systems. train a ResNet101 fine-tuned CNN to classify into two categories; landslide and non-landslide. To evaluate the 2. PROBLEM STATEMENT newly developed ResNet101 model, it is scored against commonly accepted standards such as accuracy, precision, In mountainous areas and regions of high rainfall, recall, and F1-score. Ultimately, following this evaluation, landslides are a threat to life, infrastructure and the adopted methodology incorporates the validated environment. Current landslide monitoring techniques are ResNet101 model into an online platform that allows mainly based on manual surveying, ground sensors and users to upload images and obtain real-time predictions. historical analysis. They are often expensive, timeThe adopted methodology establishes validation within consuming and limited in geographical coverage. In the proposed framework for disaster risk management addition, traditional methods are unable to offer timely through validation, scalability and usability. predictions across wide area. Satellite images provide a better geographical coverage and frequency of 5. REVIEW OF LITERATURE observations; however, their automated analysis for landslide prediction is still an ongoing process. As a result, The role of remote sensing and machine learning in the development of an accurate, scalable and automated assessing and predicting landslides has recently increased solution for the fast analysis of satellite images to provide as technology improves. Traditional methods of assessing
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 994