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 ${\delta }_1$ and footprint extraction metric $Dice$, respectively, which indicate effective improvements in the results compared to other methods.

In recent years, deep learning has been successfully applied in the field of synthetic aperture radar (SAR) image object detection. However, unlike ships and tanks targets, the oil tank targets in SAR image are usually dense and compact with more overlaps and discrete scattering centers, which greatly increases the difficulty of extracting location and structural parameters. Most of the existing methods transfer the methods suitable for natural image to the SAR image field, without considering the unique characteristics of SAR image. Therefore, in this article, we propose an improved model based on the end-to-end transformer network, which is the first model introducing transformer network to 3-D detection of oil tank targets from single SAR image. We input the incidence angle into the transformer model as a priori token. Then, we propose a feature description operator (FDO) based on the scattering centers that are used as an aid to improve the precision of predictions. In addition, we also propose a cylinder IOU as a more suitable evaluation metric for 3-D detection of oil tank. Finally, we evaluate our model on an SAR image dataset that contains SAR images from RADARSAT-2, TerraSAR-X, and GF-3 with different incidence angles. Our experiments demonstrate that our proposed model achieves the AP of 77.6% compared with 60.8% of baseline, which proves the effectiveness of the introduction of observation conditions, cylinder IOU loss (CI Loss), and the FDO based on the scattering centers in our model and is appealing for 3-D detection of oil tank.