Comprehensive Insights Into Landslide Detection using Machine Learning Approaches

Landslides are natural threats that significantly endanger human safety, infrastructure, and ecosystems, especially in regions close to forests and mountains. The model defined in the paper proposes an advanced segmentation pipeline for accurate landslide detection by applying state-of-the-art deep learning architectures and multi-modal data integration. Our approach here utilizes transformer-based segmentation models—specifically, SegFormer, a hierarchical transformer encoder paired with a lightweight MLP decoder for semantic segmentation — to produce precise picture masks that delineate landslide regions in GIS layers, SAR data, and high-resolution optical images. The system is implemented in four stages: first, a neural network is built using frameworks such as TensorFlow or PyTorch; next, it generates masks for new images; next, it compares these generated masks to the ground-truth masks using metrics like IoU, Dice Coefficient, and pixel-wise accuracy; and lastly, it determines the percentage of the segmented area compared to the full image. Extensive tests with cross-validation and data augmentation show that even under challenging conditions such as sensor noise and low lighting, our SegFormer model significantly outperforms Mask2Former.