Executive Summary : | Landslides are severe natural hazards caused by factors such as heavy rainfall, earthquakes, ground water-level variations, snow melt, and human intervention. They cause an estimated annual loss of $20 billion globally. Deep learning models and remote sensing studies have been developed to investigate landslides, and India's research and educational institutes are working towards mitigation. To detect landslides from remote sensing images (RSIs), nine deep learning models will be proposed, including U-Net, U-Net+ResNet-50, U-Net+ResNet-101, U-Net+VGG-19, U-Net+DenseNet-121, U-Net+Inception-V3, U-Net+Inception-ResNet-V3, FCN+VGG-16, and FCN+VGG-8. The output will be evaluated quantitatively in terms of precision, recall, f1-score, Matthews-correlation-coefficient, and overall accuracy. The development of this model will be significant in identifying and mapping regions vulnerable to landslides in Uttarakhand. Accurate detection of landslides will be valuable for experts, as it helps estimate the size of landslides, aiding in inventories. The model will also be useful in post-disaster analysis for fast and accurate identification of changes. In conclusion, the integration of deep learning and remote sensing is crucial for investigating landslides and improving risk management. The proposed model will be developed within the Python environment and will prove useful in identifying potential landslides in Uttarakhand, aiding in early prediction and improved risk management. |