High-Resolution Remote Sensing Dataset for Typical Disaster-Damaged Transportation Elements and Landslides in CPEC and Southeast Asia

Natural disasters such as earthquakes and extreme hydrological events often cause severe damage to transportation lifelines and trigger secondary geological disasters including landslides. The rapid and accurate acquisition of disaster information is of crucial importance to emergency rescue efforts. However, current research on intelligent remote sensing interpretation for disasters is severely constrained by the scarcity of post-disaster damaged samples and the poor generalization ability of landslide detection models in low-vegetation areas. To address these challenges, this study constructs a high-resolution remote sensing dataset for detecting damaged transportation elements and landslides, covering typical regions such as the China-Pakistan Economic Corridor and Southeast Asia. This dataset integrates three core disaster elements—damaged roads, damaged bridges and landslides—into a unified framework: to tackle the shortage of damaged samples for roads and bridges, a Stable Diffusion model based on topological constraints is proposed to generate high-fidelity synthetic images; for landslides in low-vegetation areas, a sample set for complex arid mountainous areas is established through multiple rounds of cross visual interpretation, based on high-resolution post-disaster satellite images and historical vector data. In addition, authentic post-disaster images are used to supplement the datasets of the three types of elements. After standardization processing including unified sizing and mask binarization, as well as stringent quality control, a high-quality dataset is finally formed, which contains 9102 pairs of road samples, 6061 pairs of bridge samples and 7614 pairs of landslide samples. This dataset fills the gap in high-quality annotated samples of post-disaster damaged transportation elements, supplements landslide samples in low-vegetation areas, and provides reliable multi-scenario data support for the training and validation of intelligent remote sensing interpretation algorithms for disasters and the implementation of post-disaster emergency assessment.

地震、极端水文事件等自然灾害往往会对交通生命线造成严重破坏，并引发滑坡等次生地质灾害。快速、精准获取灾情信息对于应急救援工作至关重要。然而，当前灾害智能遥感解译研究受限于灾后损毁样本稀缺，且滑坡检测模型在低植被区域的泛化能力欠佳，其发展受到严重制约。为应对上述挑战，本研究构建了一套用于交通损毁要素与滑坡检测的高分辨率遥感数据集，覆盖中巴经济走廊、东南亚等典型区域。该数据集将损毁道路、损毁桥梁与滑坡三类核心灾情要素整合至统一框架中：针对路桥损毁样本不足的问题，本研究提出一种基于拓扑约束的Stable Diffusion模型以生成高保真合成图像；针对低植被区域的滑坡样本，本研究基于灾后高分辨率卫星影像与历史矢量数据，通过多轮交叉目视解译构建了复杂干旱山区滑坡样本集。此外，本研究还采用真实灾后影像对三类要素的数据集进行补充。经统一尺寸、掩码二值化等标准化处理与严格质量管控后，最终形成高质量数据集，其中包含道路损毁样本9102对、桥梁损毁样本6061对以及滑坡样本7614对。该数据集填补了灾后交通损毁要素高质量标注样本的空白，补充了低植被区域的滑坡样本，可为灾害智能遥感解译算法的训练与验证、灾后应急评估工作的开展提供可靠的多场景数据支撑。