利用光学和SAR遥感影像进行建筑高度估计对于理解城市形态和优化城市存量空间具有重要意义。然而，现有的数据集存在诸多局限：由于样本数量较少，难以满足基于深度学习的遥感信息提取需求，样本所覆盖的区域较为有限，无法提供足够的地理多样性和空间特征代表性,特别是针对中国区域的大规模建筑高度数据集尤为缺乏。此外，数据集的开源性不足，限制了其在更广泛的研究中的应用和验证。为解决这些问题，本文构建了一个专注于建筑高度回归的数据集BHDSI，涵盖了中国62个城市的中心城区，共有5606个样本，覆盖了城市，农村等场景，是目前中国区域覆盖面积最大的建筑高度数据集。该数据集包含哨兵-1和哨兵-2的遥感影像以及建筑高度的真实值，样本大小是256×256，相比于64×64大小的数据集，为建筑高度估计研究提供了一个重要的补充选择。相比其他数据集，该数据集具有样本数量大、覆盖范围广、建筑高度分布合理等特点，能够更好地满足深度学习网络的训练需求。在此基础上，本文采用相同的深度学习网络对BHDSI数据集及其他类似数据集进行了评估，并对比了多个网络使用BHDSI数据集时在建筑高度回归任务中的表现，深入分析了各网络的优劣。研究表明，与其他数据集相比，BHDSI数据集在建筑高度回归任务中的表现更加优异。进一步分析发现，使用BHDSI数据集时，建筑高度较低的区域其估计精度相对较高。此外，U-Net解码器用于建筑高度估计网络训练能够取得更高的精度。这一数据集及实验结果为未来建筑高度估计领域的研究提供了重要的支持。

Objective The objective of this study is to address the limitations of existing datasets used for building height estimation from optical and SAR remote sensing imagery. Current datasets often suffer from small sample sizes, limited geographic diversity, and a lack of openness, making them insufficient for supporting deep learning-based remote sensing applications—especially for large-scale studies in China. Accurately estimating building heights is critical for understanding urban morphology and optimizing urban stock space, thus necessitating the development of a more comprehensive, representative, and accessible dataset. Methods To overcome these issues, this paper constructs a new dataset named BHDSI, specifically designed for building height regression tasks. The dataset comprises 5,606 samples from the central urban areas of 62 cities across China, making it the largest building height dataset for the country in terms of geographic coverage. It includes both Sentinel-1 and Sentinel-2 imagery along with true building height values, with each sample having a spatial resolution of 256×256 pixels. This provides a richer spatial context compared to existing datasets with smaller sample sizes, such as 64×64. The dataset encompasses a wide range of scenarios, including urban and rural areas, ensuring better representation of spatial features. Result Experimental evaluations demonstrate that the BHDSI dataset leads to superior performance in building height regression tasks when compared to other similar datasets, across various deep learning networks. The results also indicate that estimation accuracy tends to be higher in regions with lower building heights. Furthermore, the study finds that using a U-Net decoder structure in the network architecture contributes to higher prediction precision, highlighting the importance of decoder design in deep learning-based height estimation. Conclusion The BHDSI dataset significantly advances the field of building height estimation by offering a large-scale, diverse, and high-quality resource tailored for deep learning. Its broad coverage, balanced height distribution, and open accessibility make it better suited for training and evaluating deep neural networks than previously available datasets. The study confirms that both data quality and network architecture, especially decoder design, play vital roles in improving estimation accuracy, and BHDSI serves as a strong foundation for future research in this domain.