Large-scale Small-change Multi-modal Dataset (LSMD)
收藏资源简介:
LSMD是由安徽大学团队构建的双时相RGB-NIR多模态建筑变化检测基准数据集,聚焦大场景下的小目标变化检测难题。该数据集包含8000对高精度配准的双波段图像对,每对图像均同步采集可见光与近红外数据,空间分辨率达亚米级。数据采集严格模拟真实遥感监测场景,特别关注变化比例低于2%的稀疏变化样本,以及植被覆盖下的微小建筑变化。通过多光谱物理响应差异,为复杂城市场景中的精细变化检测提供了光谱互补性强的研究平台,可有效支持城市规划、灾害评估等应用。
LSMD is a benchmark dataset for bi-temporal RGB-NIR multimodal building change detection, developed by the research team from Anhui University, which focuses on the challenge of small-object change detection in large-scale scenes. This dataset includes 8000 pairs of high-precision registered dual-band image pairs, where each pair synchronously acquires visible light and near-infrared data, with a spatial resolution reaching the sub-meter level. The data collection strictly simulates real remote sensing monitoring scenarios, with particular emphasis on sparse change samples with a change proportion lower than 2%, as well as subtle building changes covered by vegetation. By utilizing the differences in multispectral physical responses, it provides a research platform with strong spectral complementarity for fine-grained change detection in complex urban scenarios, and can effectively support applications such as urban planning and disaster assessment.

- 1Multi-Modal Building Change Detection for Large-Scale Small Changes: Benchmark and Baseline安徽大学·计算机科学与技术学院; 安徽理工大学·公共安全与应急管理学院; 南洋理工大学·计算与数据科学学院; 陕西科技大学·人工智能联合实验室; 西安交通大学·软件工程学院; 东京大学·前沿科学研究生院 · 2026年



