Airborne array-interferometric synthetic aperture radar (array-InSAR), one of the implementation methods of tomographic SAR (TomoSAR), has the advantages of all-time, all-weather, high consistency, and exceptional timeliness. As urbanization continues to develop, the utilization of array-InSAR data for building detection holds significant application value. Existing methods, however, face challenges in terms of automation and detection accuracy, which can impact the subsequent accuracy and quality of building modeling. On the other hand, deep learning methods are still in their infancy in SAR point cloud processing. Existing deep learning methods do not adapt well to this problem. Therefore, we propose a Position-Feature Attention Network (PFA-Net), which seamlessly integrates positional encoding with point transformer for SAR point clouds building target segmentation tasks. Experimental results show that the proposed network is better suited to handle the inherent characteristics of SAR point clouds, including high noise levels and multiple scattering artifacts. And it achieves more accurate segmentation results while maintaining computational efficiency and avoiding errors associated with manual labeling. The experiments also investigate the role of multidimensional features in SAR point cloud data. This work also provides valuable insights and references for future research between SAR point clouds and deep learning.

The array interferometric synthetic aperture radar (Array InSAR) system resolves shadow issues by employing two scans in opposite directions, facilitating the acquisition of a comprehensive three-dimensional representation of the observed scene. The point clouds obtained from the two scans need to be transformed into the same coordinate system using registration techniques to create a more comprehensive visual representation. However, the two-point clouds lack corresponding points and exhibit distinct geometric distortions, thereby preventing direct registration. This paper analyzes the error characteristics of array InSAR point clouds and proposes a robust registration method for array InSAR point clouds in urban scenes. It represents the 3D information of the point clouds using images, with pixel positions corresponding to the azimuth and ground range directions. Pixel intensity denotes the average height of points within the pixel. The KAZE algorithm and enhanced matching approach are used to obtain the homonymous points of two images, subsequently determining the transformation relationship between them. Experimental results with actual data demonstrate that, for architectural elements within urban scenes, the relative angular differences of registered facades are below 0.5°. As for ground elements, the Root Mean Square Error (RMSE) after registration is less than 1.5 m, thus validating the superiority of the proposed method.

Tomographic synthetic aperture radar (TomoSAR) possesses 3D imaging capability, making it significant for building reconstruction using TomoSAR data. The reconstruction algorithm is closely related to building point cloud detection, while traditional detection methods suffer from low automation and reliance on manual configuration. This study proposes a building point cloud reconstruction method based on deep learning semantic segmentation. Initially, deep learning method is employed for end‐to‐end building point cloud segmentation, followed by point cloud reconstruction based on the segmentation results. The proposed method is simple and efficient, elevating the level of automation in point cloud processing. Experimental validation on real TomoSAR data confirms that the proposed method achieves automated and refined reconstruction of building point clouds.

Layover detection is crucial in 3D array SAR topography reconstruction. However, existing algorithms are not automated and accurate enough in practice. To solve this problem, this paper proposes a novel layover detection method that combines the complex-valued (cv) neural network and expert knowledge to extract features in the amplitude and phase of multi-channel SAR. First, inspired by expert knowledge, a fast Fourier transform (FFT) residual convolutional neural network was developed to eliminate the training divergence of the cv network, deepen networks without extra parameters, and facilitate network learning. Then, another innovative component, phase convolution, was designed to extract phase features of the layover. Subsequently, various cv neural network components were integrated with FFT residual learning blocks and phase convolution on the skeleton of U-Net. Due to the difficulty of obtaining SAR images marked with layover truths, a simulation was performed to gather the required dataset for training. The experimental results indicated that our approach can efficiently determine the layover area with higher precision and fewer noises. The proposed method achieves an accuracy of 97% on the testing dataset, which surpasses previous methods.

As a novel system, multi-dimensional space joint-observation SAR (MSJosSAR) can simultaneously acquire target information across multiple dimensions such as frequency, angle, and polarization. This capability facilitates a more comprehensive understanding of the target and enhances subsequent recognition applications. However, current research on the configuration optimization of multi-dimensional SAR systems is limited, particularly in balancing recognition requirements with observation costs. This limitation has become a major bottleneck restricting the development of MSJosSAR. Moreover, studies on the joint utilization of multi-dimensional information at the scattering center level remain insufficient, which constrains the effectiveness of target component recognition. To address these challenges, this paper proposes a configuration optimization method for MSJosSAR based on the separability of scattering mechanisms. The approach transforms the configuration optimization problem into a vector separability problem commonly addressed in machine learning. Experimental results demonstrate that the multi-dimensional configuration obtained by this method significantly improves the classification accuracy of scattering mechanisms. Additionally, we propose a feature extraction and classification method for scattering centers across frequency and angle-polarization dimensions, and validate its effectiveness through electromagnetic simulation experiments. This study offers valuable insights and references for MSJosSAR configuration optimization and joint feature information processing.