Artificial Intelligence Techniques for Landslides Prediction Using Satellite Imagery

This study presents a comprehensive approach to predicting landslides using artificial intelligence techniques applied to satellite imagery. The process begins with the collection of satellite images, which are resized to a uniform dimension to ensure consistency in analysis. Utilizing a deep learning model, specifically the ResNet101 neural network, the images are subjected to landslide classification. The ResNet101 architecture, known for its robust feature extraction capabilities, enables the model to effectively differentiate between areas prone to landslides and those that are stable. Through the training process, the network learns to identify critical features associated with landslide occurrences, leveraging the vast amount of data available from satellite imagery. The classification task is framed as a binary problem, categorizing each input image as either a "landslide" or "non-landslide." The results of this study underscore the potential of leveraging advanced artificial intelligence techniques in environmental monitoring and disaster risk reduction. By providing accurate and timely predictions of landslide occurrences, this approach can significantly contribute to improved land management practices and enhance the effectiveness of early warning systems, ultimately aiding in mitigating the risks associated with landslides in vulnerable regions.

Index Terms— Landslide classification, satellite image classification, CNN, fuzzy-based classification, Deep Leaning, landslide prediction, landcover classification