RSiDF-T: Remote Sensing image-guided Diffusion Filling with Topology regularization for SRTM Data Restoration

This is the official code of RSiDF-T: Remote Sensing Image-Guided Diffusion Filling with Topology Regularization for SRTM Restoration over Complex Terrains. We propose a conditional diffusion framework (RSiDF-T) for restoring SRTM DEM voids over complex terrains. The method adopts a coarse-to-fine two-stage reconstruction scheme. In the coarse stage, we design an optical-guided contextual attention module (OCAM) to leverage texture and structural cues from co-registered Sentinel-2 optical imagery and construct an initial elevation-structure prior for void regions. In the refinement stage, a conditional diffusion model progressively reconstructs high-fidelity elevations within gaps, assisted by a topology-aware regularization term to reduce boundary artifacts and preserve key topographic structures such as ridges and valleys.The SRTM DEM data used in this study are publicly available from the U.S. Geological Survey EarthExplorer. Sentinel-2 imagery used in this study is publicly available from the Copernicus Data Space Ecosystem.本仓库为RSiDF-T（面向复杂地形SRTM数据修复的遥感图像引导拓扑正则化扩散填充方法，Remote Sensing Image-Guided Diffusion Filling with Topology Regularization for SRTM Restoration over Complex Terrains）的官方代码实现。我们提出了一种条件扩散框架（RSiDF-T），用于修复复杂地形下的SRTM DEM（数字高程模型，Digital Elevation Model）空洞区域。该方法采用从粗到细的两阶段重建方案：在粗重建阶段，我们设计了光学引导上下文注意力模块（OCAM，Optical-Guided Contextual Attention Module），以利用配准后的Sentinel-2（哨兵二号）光学影像中的纹理与结构线索，为空洞区域构建初始高程结构先验；在精修阶段，条件扩散模型在拓扑感知正则化项的辅助下，逐步对间隙区域完成高保真高程重建，有效减少边界伪影并保留山脊、山谷等关键地形结构。本研究使用的SRTM DEM数据可从美国地质调查局地球探索者（U.S. Geological Survey EarthExplorer）公开获取；所用的Sentinel-2影像则可从哥白尼数据空间生态系统（Copernicus Data Space Ecosystem）公开获取。