In the scene of remote sensing image (RSI) recognition, it is difficult to obtain a sufficient number of samples for training all categories at once. A more realistic situation is that the recognition task occurs in an open environment, with categories gradually increasing. Additionally, due to the difficulty of collecting certain data, there are only a few samples for each new category. This leads to the problem of few-shot class-incremental learning (FSCIL), where the model learns incrementally and the number of samples for incremental classes is very small, generally only a few, while the number of samples for base classes is relatively large. To address this, this paper proposes a model framework for FSCIL of RSIs, called HDCPAA. The model is mainly divided into three parts. The first part is the feature extraction network, which is pre-trained on the base classes and then its parameters are frozen in subsequent incremental learning to alleviate catastrophic forgetting of the base classes. The second part is a fully connected layer, which transforms the prototypes of each category into quasi-orthogonal prototypes to increase the distance between the prototypes. The third part is the prototype adaptation attention module, which adaptively updates prototypes and query vectors using attention mechanisms. The training process of this module is based on the meta-learning of pseudo-incremental classes. Experiments on two popular benchmark RSI datasets, MSTAR and NWPU-RESISC45, show that our model significantly outperforms the baseline models and sets new state-of-the-art results with remarkable advantages. Our code will be uploaded at: https://github.com/lipeng144/HDCPAA .

Although deep neural networks (DNNs) have achieved good results on some common datasets of synthetic aperture radar (SAR) automatic target recognition (ATR), DNNs are opaque and difficult to interpret, which limits its practical application. In view of this, many interpretable models have been proposed in recent years. However, the previous interpretable models only represent transparent network structures and cannot be called real “interpretable” models. We think that the interpretability of a model contains two meanings: on the one hand, the decision process of the model is transparent, and on the other hand, the decision-making basis of the model should be reasonable and conform to human cognition. For this reason, we propose a new SAR image recognition model based on adversarial defense, namely SAR-AD-BagNet. Not only does it have a transparent decision-making process, but it also has a more reasonable basis for decision-making. In addition, the model also has high SAR image recognition accuracy and strong adversarial robustness.

Wideband interference (WBI) is detrimental to high-resolution radar due to its high power and wide frequency occupancy. In this study, a deep learning (DL) method is proposed to predict the time–frequency (TF) feature of WBI and applied to cognitive radar high-resolution range profile (HRRP) estimation. Specifically, by performing short-time Fourier transform (STFT) on the WBI signal collected in the past and using a sliding window, a series of WBI TF figures is generated. A long short-time memory (LSTM) network is then used to learn the spatiotemporal (ST) correlation of these TF figures, thus predicting the WBI TF feature in the future, based on which, a cognitive method is used for target HRRP estimation with reduced influences of WBI. Numerical results demonstrate the effectiveness of the proposed methods.

There are a number of targets in space that are similar in shape, but they pose different threat levels. It is necessary to discriminate these targets for the sake of space security. The unique characteristics of detailed micro-Doppler (MD) signatures can be used to classify micromotion forms, such as spinning, precession, nutation, wobbling, and tumbling. In this letter, we design a novel processing flow for classifying five types of micromotion under low signal-to-noise ratio (SNR) condition. We start from the range map and successively perform noise-level estimation, adaptive filtering, and multithreshold segmentation to obtain an ideal binary mask. The clean range maps are used as the final input to an eight-layer convolutional neural network (CNN) for training, validation, and testing. The experiment results show that our method achieves more than 80% classification accuracy for micromotion in the case of −10 dB, which is higher than the existing popular methods.

Microwave imaging with large rotating angle and sparse sampling is an attractive approach to obtain the high-resolution target image with reduced radar resource. However, the popular imaging methods, e.g., Range-Doppler (RD), back projection (BP), and sparse recovery (SR), are difficult to deal with large rotating angle and sparse sampling simultaneously. In recent years, deep learning (DL) has been widely studied and been successfully used to handle the problems in computer vision. However, since most existing DL networks are put forward for the real visual image and a large amount of data is essential for network training, DL cannot be directly used to process the complex and sparse target echo for microwave imaging. In this article, a new learning imaging framework is proposed and a model-driven learning imaging network (MDLI-Net) is built for high-resolution microwave imaging with large rotating angle and sparse sampling. In the proposed framework, the electromagnetic scattering model is used to generate the training data efficiently, and the sparse microwave imaging theory is applied to guide the design of the deep imaging network. By inputting the 2-D sparse complex-valued target echo, the trained MDLI-Net can output the high-resolution and focused target image efficiently. The effectiveness of the proposed learning imaging method is validated by experiment results with both simulated and real data.

A cognitive antenna selection strategy for multi-in multi-out (MIMO) imaging radar is proposed in this article. The basic idea of our strategy relies on the dynamic selection of the antenna location according to the feedback information to enhance the image quality and reduce the computational burden simultaneously. The aim of our strategy is to indirectly minimize the mean squared error associated with the amplitudes and positions of the strong scattering centers of the target through the frame potential. Specifically, it is assumed that the imaging process is initially performed via a conventional uniform linear array/random sparse array of collocated MIMO radar; hence, based on the collected data, an initial image of the target is derived (perception). Then, the accuracy of low-resolution images is enhanced progressively according to the cognitive paradigm via a specific antenna location selection at the next transmission (action). Benefit from the enhanced accuracy, the support area of targets can be estimated to reduce the dimension of the undersampling matrix and, finally, the computational burden in the subsequent high-resolution image reconstruction process is reduced. The simulation results highlight the capabilities of our cognitive approach to provide more interesting benefits in imaging than the random selection strategy and demonstrate the enhanced imaging performance of cognitive sparse MIMO array under the condition of limited antennas and noise.

The problem of waveform separation for the multiple-input multiple-output (MIMO) imaging radar is considered in this article. In the conventional match filtering (MF)-based waveform separation methods, the cross-correlation noise among different waveforms is inevitable and may seriously reduce the performance of the MIMO radar, especially in target imaging, where a large number of transmissions are required. In this article, a cognitive waveform separation method is proposed based on the orthogonality of the frequency-stepped signals in the Doppler domain. The basic idea of the proposed waveform separation method is to extract different spectrums that occupy different Doppler coverages after interpulse phase modulation. In particular, the mix operation before filtering is conducted due to its compression effort to hold all the Doppler spectrums in one period. Depending on the prior knowledge, the target characteristics, and the imaging resolution requirement, a cognitive optimization model considers both the waveform separation and the target imaging to obtain the most suitable transmitting parameters for different targets. The inner relationships between the transmitting parameters, waveform separation, and radar imaging are fundamental to transmitting the parameter optimization. The perception-action cycle of cognition is achieved via the quantum genetic algorithm (QGA). The simulation results show the superiority of the proposed method with respect to the MF-based separation methods in the MIMO radar imaging.

Multiple-input multiple-output (MIMO) radar imaging is a new technique to obtain the radar image of aerospace targets. Orthogonal waveform design is one of the important issues for MIMO radar imaging. However, the fully orthogonal waveforms in the same frequency and with the arbitrary time delay do not exist in practice. Thus, the imaging result using nonorthogonal waveforms based on matched filtering (MF) method is usually unsatisfactory if further processing like digital beam forming (DBF) is not used. Sparse recovery (SR) method is possible to restrain the mutual interference of nonorthogonal waveforms by exploiting the sparsity of targets and improve the imaging quality. In this article, waveform design issue in SR-based MIMO imaging method is studied. The difference in the designs of waveforms in MF method and SR method is discussed. Based on requirements analysis, a comprehensive optimization model is built for waveform design and the existing cycle algorithm (CA) is modified to solve the model. Considering the fact that the target scene is always changing, waveforms should be adjusted along with the dynamic scene. Therefore, an adaptive waveform optimization method is further proposed based on the cognition of target scene. The dimension of SR model is reduced and the waveforms are optimized according to the cognitive target length. Moreover, based on the reconstructed target range profiles, transmitting waveforms together with recovery algorithm are further optimized to match the target better. Simulation results show that the waveforms after optimization are better than the nonoptimized waveforms and the proposed adaptive optimization method is valid and robust for the dynamic target scene.

In actual condition, array elements deficiency or transmission errors lead to incomplete data, which is called sparse aperture (SA) data. In inverse synthetic aperture radar (ISAR) imaging, this large‐gaped data produces poor‐quality ISAR images when using traditional range–Doppler algorithm. Recently, imaging algorithms based on compressed sensing (CS) theory alleviate this problem effectively because CS theory indicates that sparse signal can be reconstructed from incomplete measurements. However, the basis mismatch problem in CS‐based algorithms may degrade the ISAR image. In this study, a reweighted atomic‐norm minimisation (ANM) (RAM)‐based imaging method is proposed. RAM is a gridless sparse method, which can enhance sparsity and resolution. RAM formulates an optimisation problem and iteratively carries out ANM with a sound reweighting strategy. By reformulating the RAM as a semi‐definite programme, the echoes with full aperture (FA) are reconstructed from SA data. After that, ISAR imaging with the reconstructed FA data is achieved via the conventional azimuth compression method. Simulated and real data results demonstrate the effectiveness and superiority of the proposed method.

Multiple‐input‐multiple‐output (MIMO) radar is attractive in moving targets imaging, which is able to solve the difficulty of complex motion compensation. In this article, a three‐dimensional imaging method using bistatic MIMO radar is proposed. Compared with monostatic MIMO radar, the bistatic system can provide the complementary information, and the imaging process is not the same as the monostatic case. Furthermore, considering the image sparsity and the spatial limitation of radar targets, a fast smoothed L0 norm algorithm is proposed to achieve the high resolution in cross‐range directions with limited antennas. The experimental results demonstrate the validity and efficiency of the proposed method.

Multiple-input multiple-output (MIMO) radar imaging is a new technique which may solve the motion compensation problem in inverse synthetic aperture radar (ISAR). However, the imaging result in MIMO radar using matched filtering is usually poor, since the waveforms with the same frequency cannot be fully orthogonal. Sparse signal recovery has the potential to restrain the mutual interference of nonorthogonal waveforms by exploiting the sparsity of targets. However, because the range profile is not as sparse as the 2-D or 3-D image, the sparse recovery result of target range profiles is usually unsatisfactory. In this paper, a joint-block sparsity of range profiles is explored and exploited to improve the range profile quality. And then, the 2-D target image is recovered from the refined range profiles. Furthermore, a robust joint-block sparse recovery algorithm is proposed. The ascent searching direction, the parameter selection method, and the computational complexity of the proposed algorithm are also discussed. Simulation results show that the proposed algorithm is superior to algorithms which just consider sparsity, block sparsity, or joint sparsity. And the quality of the simulated MIMO radar images and real data ISAR images obtained using the new imaging method is better than that of the conventional correlation method and sparse signal recovery method.

Compressive sensing (CS) has been introduced into inverse synthetic aperture radar (ISAR) imaging with partial measurements. However, in the case of transmitting sparse frequency-stepped chirp signal (FSCS), the CS-based method will produce an irregular range cell migration (IRCM) problem in the recovered high-resolution range profiles (HRRPs). The IRCM is induced by the basis mismatch problem in CS, and it will degrade the ISAR image. To obviate the IRCM, an atomic norm minimization (ANM) method is proposed in this letter. By reformulating the ANM as a semidefinite program (SDP), the echo with full FSCS can be recovered using off-the-shelf SDP solvers. Thus, HRRPs without IRCM can be achieved via the conventional inverse fast Fourier transform. As a result, an improved ISAR image will be obtained. Real data results demonstrate the advantages of the proposed method over the CS-based and matrix completion-based methods.