Search

Extracting buildings in urban scenes from remote sensing images is crucial for the construction of digital cities, urban monitoring, urban planning, and autonomous driving. Traditional methods generally rely on shadow detection or stereo matching from multi-view high-resolution remote sensing images, which is cost-intensive. Recently, machine learning has provided solutions for the estimation of building heights from remote sensing images, but challenges remain due to the limited observation angles and image quality. The inherent lack of information in a single modality greatly limits the extraction precision. This article proposes an advanced method using multi-source remote sensing images for urban building height estimation, which is characterized by multi-level cross-fusion, the multi-task joint learning of footprint extraction and height estimation, and semantic information to refine the height estimation results. The complementary and effective features of synthetic aperture radar (SAR) and electro-optical (EO) images are transferred through multi-level cross-fusion. We use the semantic information of the footprint extraction branch to refine the height estimation results, enhancing the height results from coarse to fine. Finally, We evaluate our model on the SpaceNet 6 dataset and achieve 0.3849 and 0.7231 in the height estimation metric δ 1 and footprint extraction metric D i c e , respectively, which indicate effective improvements in the results compared to other methods. [ABSTRACT FROM AUTHOR]

Accurately estimating building heights is crucial for various applications, including urban planning, climate studies, population estimation, and environmental assessment. However, this remains a challenging task, particularly for large areas. Satellite-based Light Detection and Ranging (LiDAR) has shown promise, but it often faces difficulties in distinguishing building photons from other ground objects. To address this challenge, we propose a novel method that incorporates building footprints, relative positions of building and ground photons, and a self-adaptive buffer for building photon selection. We employ the Ice, Cloud, and Land Elevation Satellite 2 (ICESat-2) photon-counting LiDAR, specifically the ICESat-2/ATL03 data, along with building footprints obtained from the New York City (NYC) Open Data platform. The proposed approach was applied to estimate the heights of 17,399 buildings in NYC, and the results showed strong consistency with the reference building heights. The root mean square error (RMSE) was 8.1 m, and for 71% of the buildings, the mean absolute error (MAE) was less than 3 m. Furthermore, we conducted an extensive evaluation of the proposed approach and thoroughly investigated the influence of terrain, region, building height, building density, and parameter selection. We also verified the effectiveness of our approach in an experimental area in Beijing and compared it with other existing methods. By leveraging ICESat-2 LiDAR data, building footprints, and advanced selection techniques, the proposed approach demonstrates the potential to accurately estimate building heights over broad areas. [ABSTRACT FROM AUTHOR]

Extracting buildings in urban scenes from remote sensing images is crucial for the construction of digital cities, urban monitoring, urban planning, and autonomous driving. Traditional methods generally rely on shadow detection or stereo matching from multi-view high-resolution remote sensing images, which is cost-intensive. Recently, machine learning has provided solutions for the estimation of building heights from remote sensing images, but challenges remain due to the limited observation angles and image quality. The inherent lack of information in a single modality greatly limits the extraction precision. This article proposes an advanced method using multi-source remote sensing images for urban building height estimation, which is characterized by multi-level cross-fusion, the multi-task joint learning of footprint extraction and height estimation, and semantic information to refine the height estimation results. The complementary and effective features of synthetic aperture radar (SAR) and electro-optical (EO) images are transferred through multi-level cross-fusion. We use the semantic information of the footprint extraction branch to refine the height estimation results, enhancing the height results from coarse to fine. Finally, We evaluate our model on the SpaceNet 6 dataset and achieve 0.3849 and 0.7231 in the height estimation metric δ1 and footprint extraction metric Dice, respectively, which indicate effective improvements in the results compared to other methods.

In recent years, limited surface space and restricted urban expansion have become increasingly prominent. The development of underground spaces has emerged as an effective solution, offering a new dimension for urban growth and sustainability. However, due to the non-renewable and costly nature of underground space (UGS), it should be properly planned before development to avoid waste of resources and economic loss. In this study, the potential of UGS development based on the influence of surface buildings was evaluated by using the high-resolution satellite images of Gaofen-2 (GF-2) in Qingdao, Shandong Province, China. Firstly, the DeepLabv3+ network was lightened to enhance the training efficiency and building extraction accuracy. The building shadows were then extracted by an improved shadow measurement model, and used to estimate the height of extracted buildings based on the imaging mechanism and information of the remote sensing image. Finally, the building height (BH) was converted into the depth of influence on the UGS, and the development potential of the UGS was then estimated. The experimental results show that the UGS within 0–10 m in Qingdao city has been extensively occupied. However, the geological layers between 10–30 m and 30–50 m contain large continuous areas that have not been utilized, indicating the greatest potential for resource development within the city. [ABSTRACT FROM AUTHOR]

Building height, as an essential measure of urban vertical structure, is key to understanding how urbanization is reshaping inner-city characteristics, particularly in developing countries. However, estimating building height in urban environments remains challenging. Building height estimation with physical model-based feature approaches and machine learning approaches are limited by a constrained large-scale application capability and the lack of physical significance, respectively. In this study, we proposed a two-step method to estimate building height in spatially heterogeneous urban areas by integrating the merits of machine learning approaches and physical model-based features, together with spatial contextual information. First, we trained a block-level machine learning model on Hangzhou block units to estimate average block-level building height as spatial contextual information. Second, we trained a building-level machine learning model to estimate the final building height of Hangzhou with the estimated spatial contextual information and additional physical model-based features, including radar look angle, building wall orientation, the length of the building, and dielectric constants of the building wall. Our results showed that the proposed method can largely improve the performance of building height estimation, with an overall R2 and RMSE of 0.76 and 6.64 m, respectively. Incorporating physical model-based features and spatial contextual information reduced model RMSE by 32 %. Compared with existing methods, our proposed model demonstrated a better accuracy performance and improved capability in addressing the prevailing overestimation of low-rise buildings and the underestimation of high-rise buildings in highly heterogeneous urban areas.