Underground Target Extraction by Local Entropy Feature of Voxelized 3D Ground Penetrating Radar

To help protect communities and historic routes, we mapped where landslides are most likely along the Ancient Qin-Shu Roads. We combined 1,407 recorded landslides with data on terrain, land cover, roads, rainfall, and ground stability, then compared six methods. The best method captured 91.8 percent of past landslides in just 13.5 percent of the area. Mapping five levels directly created clearer priority zones than a two-group approach, supporting more focused monitoring and prevention.