Deep learning has offered new ideas in SAR ship target recognition. Although many methods improve the recognition performance through the improvement of loss function and migration of deep networks, scattering features as the important intrinsic features of SAR targets, need to be considered in the SAR ship recognition tasks. To introduce the scattering features into the deep network and characterize the features of ship targets more comprehensively, a multi-scale global scattering feature association network (MGSFA-Net) for SAR ship target recognition is proposed in this paper. In the network, the SAR ship target is firstly separated from the background by fine target segmentation. Then, the scattering centers (SCs) of ship targets are extracted and converted to local graph structures based on the $k$ -Nearest Neighbors (KNN) algorithm. The local graph structures are associated by the scattering center feature association (SCFA) module and enhanced by the multi-scale feature enhancement (MSFE) module to produce the multi-scale global scattering features. Moreover, the deep features of the targets are extracted by the multi-kernel deep feature extraction (MKDFE) module to characterize the high-dimensional information. Finally, the scattering features and deep features are fused by weighted integration to enrich the diversity of features. The experimental results on the FUSAR-Ship and OpenSARShip dataset show that, the MGSFA-Net can significantly improve the recognition performance, even on a few-shot condition with the accuracy increasing over 2%-3%. The feature distribution and visualization show the effectiveness of the MGSFA-Net to characterize the multi-scale global scattering association features.

In practical SAR target detection applications, it is often encountered that the training and testing data come from different SAR sensors, leading to a decline in SAR target detection performance. Although domain adaptation methods can achieve model generalization through feature transfer, the change of scattering characteristics for the same target and the difference of feature distribution, caused by cross-sensor, cannot be ignored in SAR images. It is inevitable to lead to the escalation of the offset in the bounding box regression and deviation of the feature alignment. To address these issues, a cross-sensor SAR image target detection method based on dynamic feature discrimination and center-aware calibration is proposed. Based on the domain adaptation framework, initially, a Dynamic Feature Discrimination Module (DFDM) is introduced to address the exacerbated offset in the regression. A bidirectional spatial feature aggregation mechanism is employed to aggregate features in both horizontal and vertical directions and a multi-scale structure is adopted to enhance the scattering and semantic features, which can dynamically constrain the target position while improving target discrimination capability. Then, the Center-Aware Calibration Module (CACM) is designed to address the alignment deviation in feature transfer. The target salience relationship is modeled based on the distance between different positions and the target center to suppress background clutter interference. The perception center of the target is focused by combining the centerness map and classification map, which can calibrate the domain-invariant features and alleviate misalignment. Finally, the proposed method is tested on two datasets, MiniSAR and FARAD, and compared with the latest domain adaption methods. Both mAP and F1 values have improved by more than 6%-20%, verifying the effectiveness of the proposed method.

Urban area extraction using polarimetric synthetic aperture radar (PolSAR) data has become a potential mean to urban studies because it holds the promise that the radar returns from specific scattering characteristics may be emphasized. Due to the high variability of urban land-scape and the existing misdetection of buildings as vegetation, urban area extraction is still a challenging problem. In this paper, an eigenvalue-based urban area extraction method is proposed. First, similar to the entropy/anisotropy plane, a two-dimensional RVI/PA plane is put forward to construct the extractor of buildings with small orientation angles. Second, coupled with the parameters, a robust extractor is introduced to elevate the scattering characteristics of buildings with large orientation angles but to suppress those of others. Finally, data-driven thresholds are investigated and ascertained for the extractors, thus urban areas are extracted. In addition, a change detection-based prescreening method is applied to refine the extraction result. The performance of the proposed method is demonstrated and validated with spaceborne and airborne fully PolSAR data over different test sites. The outputs show that the proposed method provides an overall accuracy of over 90%, as well as better visual results of the extracted buildings.

Downward-looking linear array 3-D synthetic aperture radar (SAR) has attracted increasing attention in the field of radar imaging. As widely reported, the volume of data can be significantly reduced by a random sparse linear array. However, the 2-D under-sampled azimuth-cross-track data brought by the sparse linear array will produce high-level side-lobes, as well as the aliasing and the false-alarm targets. To deal with those problems, this paper introduces a recently developed theory, matrix completion (MC). The new theory could recover a matrix with a small subset of known elements of the matrix. It is founded on the assumption that the matrix is essentially low rank. For downward-looking 3-D SAR with a random sparse linear array, the received 3-D data can be treated as a series of uncorrelated 2-D matrices by the separated channel process. First, range compression can be realized by means of pulse compression. Then, the sets of the 2-D under-sampled azimuth-cross-track matrix can be completed into a full-sampled one via MC trick. The resulting 3-D images can be focused by synthetic aperture technique along the azimuth direction and beamforming operation along the cross-track direction, with the recovered full-sampled matrix. The proposed algorithm achieves high resolution and low-level side-lobes with the acceptable computational cost and memory consumption. It is verified by several numerical simulations and multiple comparative studies on real data. The experimental results clearly demonstrate the imaging performance across different under-sampling rates and signal-noise rates.

For downward-looking linear array 3-D synthetic aperture radar (SAR), the azimuth and cross-track resolution are unacceptable due to the length limitation of synthetic aperture and linear array. Hence, superresolution reconstruction algorithms are desired. Since the signal to be reconstructed is sparse on the 2-D azimuth-cross-track plane, it is quite suitable to apply the compressive sensing theory to obtain the images. The existed imaging algorithms for downward-looking linear array 3-D SAR are based on 1-D compressive sensing, which could bring the couple effect between different directions. To solve this problem, a novel 3-D imaging algorithm based on 2-D compressive sensing is proposed in this paper. Instead of converting the sparse reconstruction of 2-D matrix signals to the sparse reconstruction of 1-D vectors, the proposed algorithm directly reconstructs the 2-D sparse signals on overcomplete dictionaries with separable atoms. It not only provides superresolution performance, but also reduces the storage of data acquisition and processing. Furthermore, a definition of joint sparse sampling strategy is given to reconstruct the measurement matrices for further improving the computational efficiency of the imaging algorithm. Moreover, in order to investigate the limits of the proposed algorithm, the theory analysis of Cramér-Rao bound is derived and compared with the standard deviation. Finally, numerical simulations under the noise scenarios and the principle prototype experiment on real data are shown to demonstrate the validity and the limits of the proposed algorithm.