Heng Zhang

Autofocus plays a pivotal role in synthetic aperture radar (SAR) imaging, which is particularly crucial for high-resolution applications. Traditional autofocus methods face inherent limitations, such as reliance on strong isolated scatterers within the scene or prohibitive computational complexity, making them inadequate for the practical demands of the rapid and batch processing of massive SAR imagery. Although recent deep learning (DL)-based methods have improved focusing efficiency, they predominantly rely on predefined high-order polynomial models for phase error estimation, which inevitably introduces parameter redundancy, restricts the model’s adaptability to complex phase errors, and is highly prone to inducing severe oscillations at the aperture boundaries. To overcome these limitations, this article proposes DeepAutofocus, a non-parametric deep learning framework for SAR autofocus. First, we design a physics-inspired subaperture correlation Transformer architecture. A multiscale convolutional neural network (CNN) is utilized to extract local spatial features, which are then transformed into serialized subaperture image patches. Subsequently, the self-attention mechanism of the Transformer is employed to capture the long-range dependencies across subapertures. Second, we introduce a data-driven cubic B-spline reconstruction strategy that directly estimates phase errors without depending on predefined high-order polynomial models, thereby eliminating boundary oscillations while maintaining high flexibility. Extensive experiments on both simulated and real measured satellite datasets demonstrate that, compared to traditional autofocus algorithms and existing DL-based methods, the proposed DeepAutofocus achieves superior focusing performance under various complex phase error conditions.

Recent learning-based methods for building height estimation from a single SAR image have shown promising performance. However, existing methods are constrained by limited model learning capacity and their dependence on large-scale SAR-LiDAR paired datasets. To address these issues, this paper proposes a tiny-dataset scale-consistent diffusion framework, which comprises the following three components. i) A Stable Diffusion-based model is fine-tuned using just one GaoFen-3 SAR image to achieve task transfer. Dedicated data augmentation strategies are introduced and validated through ablation studies. ii) A novel and fast height ground truth pipeline is designed to generate radar-domain building facade points from sparse LiDAR data, with vegetation effects explicitly removed. Experimental analysis reveals that removing vegetation leads to improved building height estimation accuracy. iii) A geometry-driven transformation module is introduced to ensure a consistent height scale, making it universally applicable across images. Experiments on GaoFen-3 Spotlight SAR data demonstrate that the proposed framework, trained on only one image, outperforms current UNet- and Transformer-based baselines. Cross-sensor evaluation on TerraSAR-X further indicates that the proposed framework has a certain degree of generalization ability.

The Sentinel-1 (S1) satellites, operating at C-band in Terrain Observation by Progressive Scans (TOPS) mode, provide unprecedented opportunities for measuring large-scale tectonic motions using Interferometric Synthetic Aperture Radar (InSAR). However, the azimuth-varying Doppler centroid imposes stringent coregistration requirements, for which the Enhanced Spectral Diversity (ESD) method is typically employed to achieve the required high-precision azimuth coregistration. In practical applications, however, the ESD method is inevitably affected by various disturbances, especially by pronounced large-scale and non-stationary azimuth deformation. Such disturbances not only degrade the coregistration accuracy of the ESD method but also cause phase jumps at burst edges, thereby complicating the subsequent phase unwrapping. To address these issues, this study proposes an improved coregistration strategy that integrates the Local Iterative Pixel Offset Tracking (LI-POT) and Weighted ESD (WESD) to mitigate the effects of azimuth deformation and enhance the robustness of the ESD method. Furthermore, the Coherence Gradient Consistency (CGC) indicator is introduced to embed the proposed method into the standard processing workflow, enabling automated selection of appropriate coregistration strategies. The effectiveness of the proposed method was validated using S1 TOPS repeat-pass data acquired over four representative regions. By applying the proposed method, the phase jumps observed in the conventional processing were effectively corrected, demonstrating its superior accuracy and stability in complex deformation scenarios.

Ground deformation is a major geohazard in many urban areas, requiring reliable monitoring and forecasting for hazard mitigation. Although Multi-Temporal InSAR enables high-resolution deformation monitoring, most prediction approaches rely on single-point modeling and fail to exploit spatial dependencies within deformation fields. This study proposes a spatiotemporally synchronous prediction framework for large-scale InSAR deformation fields, integrating sequence preprocessing, spatiotemporal modeling, and deformation pattern analysis. First-order differencing reduces sequence non-stationarity, while a patch-based encoder-decoder structure preserves spatial topology during dimensionality reduction. The core prediction model, built on PredRNNv2, captures the long-term spatiotemporal evolution of InSAR deformation sequences. In addition, independent component analysis (ICA) combined with K-means clustering identifies dominant deformation patterns and their geological associations. The framework is evaluated using synthetic datasets simulating multiple deformation mechanisms and Sentinel-1 InSAR time-series data over the Beijing Plain from 2015 to 2025. Results show that the model accurately captures deformation evolution and identifies transitions associated with groundwater regulation. These findings demonstrate the potential of deep spatiotemporal learning for large-scale InSAR deformation prediction and geohazard mechanism interpretation.

Novel view synthesis of synthetic aperture radar (SAR) imagery is constrained by inherent properties such as view-dependent anisotropic scattering, occlusion, and layover. These constraints lead to slow convergence and limited physical interpretability in existing approaches. To overcome these bottlenecks, we propose the SAR-3DGS framework. SAR-3DGS is a differentiable rendering pipeline for novel view synthesis and 3D reconstruction from SAR data. The proposed SAR-3DGS framework establishes a forward rendering model from 3D scenes to SAR images by integrating two key modules: Scattering Splatting, driven by slant-range projection, and Shadow Splatting, which models radar beam occlusion effects. A geometry point cloud extraction algorithm based on local density analysis filters out outlier points to extract the target's three-dimensional geometry. Experiments on both simulated and real datasets comprehensively evaluate SAR-3DGS's capabilities in novel view synthesis and 3D reconstruction. Results demonstrate that, for novel view synthesis, SAR-3DGS surpasses state-of-the-art baselines and achieves markedly reduced training time. For 3D reconstruction, SAR-3DGS successfully recovers the target's 3D point cloud from multi-view SAR images. Future work will extend the method to efficient 3D reconstruction from spaceborne multi-view SAR imagery.

The azimuth multi-channel (AMC) scanned-beam synthetic aperture radar (ScanSAR) can achieve high resolution and ultrawide-swath imaging. An inter-channel phase error estimation method for AMC ScanSAR is studied in this paper. First, the signal is dechirped and coarse-focused in the range-Doppler domain. Second, a linear regression is applied to estimate the post-dechirp spectral folding number (PDSF-number). Finally, phase errors are estimated based on the PDSF-number and the focused peak values. The proposed method can accurately estimate the phase errors and restore the image quality with low computational cost and high robustness, which is verified by experiments.