Landslide monitoring: DenseNET and image segmentation techniques to classify type and compute run-out displacement and hazard area

Exploring technologies and advancements in AI is done for reaching the heights that are difficult for humans to do manually. In landslide disaster management, current approaches often emphasize real-time forecasting using sensor-based environmental data,yet such systems can be hindered by limited availability, resolution, or timeliness during actual events. To overcome these limitations, our study introduces a post-event, image-driven framework that delivers rapid and reliable analysis of landslide dynamics using satellite or drone imagery. The proposed system performs an automated reconnaissance-style assessment, achieving 99.5% accuracy in identifying landslide characteristics and 92% accuracy in occurrence-related inference. It further estimates run-out displacement (RMSE: 0.222km) and impacted area (RMSE: 0.408km2), providing essential spatial metrics for disaster response and laying the groundwork for future integration of velocity and momentum estimation. By significantly reducing interpretation time, this approach enhances situational awareness and sets the foundation for future integration of momentum and velocity estimations. With real-time deployment, such systems could play a transformative role in early warning and rapid hazard mitigation efforts.