A multi-scale factor feature fusion modeling method for landslide susceptibility mapping

Landslide susceptibility mapping (LSM) is a critical technique for geological hazard risk assessment, aiming to quantitatively evaluate and spatially delineate potentially susceptible areas. Recently, convolutional neural networks (CNNs) and Transformer architectures have been widely applied in remote sensing and geospatial contexts, significantly enhancing automatic feature extraction capabilities. However, single-model approaches often struggle to capture fine-grained local characteristics and long-range spatial dependencies in landslide-prone regions, limiting their representation of multi-scale and irregular spatial structures. To address this challenge, we propose a progressive local-to-global synergy network (PLGS-Net), which integrates the local spatial perception of CNNs with the global dependency modeling of Transformers to achieve hierarchical, collaborative representation of landslide susceptibility features. Empirical experiments conducted in a typical landslide-prone region of Ya'an, Sichuan Province, demonstrated that PLGS-Net outperforms conventional CNNs, pure Transformers, and other mainstream deep learning models in classification accuracy, generalization ability, and regional adaptability, providing an efficient and practical approach for landslide susceptibility assessment in complex terrains.