In recent years, impressive progress has been achieved in synthetic aperture radar (SAR)-based automatic target recognition (ATR) with the development of deep learning. In practice, a complete training SAR image dataset in all target-aspect domains is limited available for one measurement. When SAR images in the unseen aspect domains are newly acquired, direct retraining of the trained SAR-ATR models with them may lead to a significant performance decline for the seen aspect domains. In this article, we propose an aspect continual recognition model (ACRM) to address the learned feature forgetting problem when SAR images with different target aspects come sequentially in the real-world SAR-ATR. Considering the abundant variations of SAR images with target aspects, we introduce the Bayesian probabilistic frame to improve the model’s generalization of characterizing the varied target features across different aspects. To acquire a better solution for the posterior probability of the model parameters, we integrate an online Monte Carlo variational inference into the deep neural network in the ACRM. Furthermore, to mitigate the accumulation of estimation errors caused by the repetitive approximations in inference, we leverage the coreset method by retaining a small subset of important samples from previous tasks as a coreset. We conduct extensive experiments on the MSTAR and FUSARship datasets. Compared with a variety of baseline algorithms in continual learning, our methods exhibit excellent SAR-ATR performance and robustness, when the SAR images from different target aspects are acquired sequentially.

In real-world synthetic aperture radar (SAR)-based automatic target recognition (ATR), variations in the target-aspect lead to differences in the distribution of target’s scattering points, which will affect the model’s recognition performance, if the training SAR images are incomplete among target-aspect. Traditional deep learning methods for SAR-based recognition often suffer from catastrophic forgetting when online trained on SAR images from different target-aspect domains. To equip SAR-ATR models with the capability of recognizing SAR images online from subsequent target-aspect domains and retaining previously learned knowledge with minimal forgetting, we propose a target-aspect hard attention continual learning (THAT-CL) method, which applies a hard attention mechanism through embedding the indexes of different target-aspect recognition tasks as vectors in each network layer to memorize information of different tasks. By dynamically scaling network weight gradients, we ensure that weights containing more task-specific information undergo smaller updates, while weights with less relevant information experience larger updates. We evaluate THAT-CL using the moving and stationary target acquisition and recognition (MSTAR) dataset. Comparative analysis against other methods demonstrates that the network with THAT-CL achieves higher average accuracy of 93.58% and lower forgetting rate of 3.01%. The results highlight the excellent recognition capability of THAT-CL in generalizing across different SAR image target recognition tasks with varying target-aspects.

Non-reciprocal antenna beam patterns are promising to be utilized in asymmetric massive multiple-input multiple-output (MIMO) systems for future sixth-generation communications. The inconsistency of uplink (UL) and downlink (DL) channels makes channel modeling in this scenario challenging. In this paper, a novel geometry-based stochastic model (GBSM) is proposed for non-reciprocal terahertz (THz) channels. A directional effective scatterer generation algorithm is designed to depict the inconsistency of bidirectional propagation conditions. The correlation function between UL and DL is derived and analyzed, which validates the ability to characterize the non-reciprocal channels. To mimic THz propagation features, molecular absorption and diffuse scattering are introduced to the model, which is verified by measured data. In addition, the non-stationarities in space, time, and frequency domains are characterized, respectively. Statistical properties are compared between analytical and simulation results, and good agreements are shown. Finally, the accuracy of the model is verified by comparing with the ray tracing data.

Millimeter-wave (MMW) imaging techniques have been widely used in the public security industries for their under-controlled privacy concerns and no health hazards. However, since MMW images are low resolution and most objects are small, reflection-weak, diverse, suspicious object detection in the MMW images is a very challenging task. This paper develops a robust suspicious object detector for the MMW images based on the Siamese network integrated with the pose estimation and image segmentation, which estimates the coordinates of human joints and segments the complete human images into symmetrical body part images. Unlike most existing detectors, which detect and recognize suspicious objects in MMW images and require a complete training set with correct annotations, our proposed model aims to learn the similarity between two symmetrical human body part images segmented from the complete MMW images. Furthermore, to decrease the misdetection caused by the restricted field of view, we further fuse the multi-view MMW images observed from the same person by designing a decision-level fusion strategy and feature-level fusion strategy based on the attention mechanism. Experimental results on the measured MMW images show that our proposed models have favorable detection accuracy and speed in practical application and thus prove their effectiveness.

In recent years, deep convolutional neural networks (CNNs) pose superior synthetic aperture radar target recognition (SAR-TR) performance. However, CNN-based SAR classifiers would be vulnerable to adversarial attack (AA) when strong nonlinearity of CNN is contrapuntally utilized by AA. The AA can cause a CNN classifier to produce erroneous predictions with extremely high confidence by injecting a tiny adversarial perturbation to the input SAR images. In this letter, an accelerated SAR-TR AA algorithm is proposed named Fast C&W. We introduce a well-trained deep encoder network to replace the process of searching for the optimal perturbation of the input SAR image iteratively in the vanilla C&W algorithm. In this way, an adversarial perturbation can be generated much faster through the rapid forward mapping during an attack. Meanwhile, as a feature extraction network, the encoder network can learn the separable data region by optimizing the attack loss function. Through the encoder network, the added perturbation energy can be mainly concentrated on a region of target instead of background clutter area. This property would be of advantages in the perturbation location control in an SAR image. In the experiments, we use the proposed AA algorithm to interfere with the deep CNN-based high-accuracy SAR-TR model trained on the moving and stationary target acquisition and recognition (MSTAR) data set, which demonstrates its excellent effectiveness and thousands of times of efficiency improvement.

The higher bandwidth inverse synthetic aperture radar (ISAR) can obtain the higher resolution radar images, which can provide more target information and help improve radar target detection and recognition. It is essential to study how to achieve a precise high-resolution (HR) ISAR image utilizing limited measurement echoes. The existing neural-network-based ISAR imaging methods extract features only from limited measurement echoes, and the common features in HR ISAR images are not utilized sufficiently, which limits the imaging performance improvement. Moreover, in their loss functions, there are no explicit constraints on the correct recovery of strong scattering points, which are important in reflecting the target characteristics. In this letter, we propose a conditional probabilistic generative model to achieve the HR ISAR imaging. By optimizing the well-designed Kullback–Leibler (KL) divergence between conditional prior and approximate posterior probability distribution in the loss function, the common features contained in training HR radar images can be learned, and a suitable prior probability distribution for the latent variable can be obtained. To accurately recover the positions and relative amplitudes of strong scattering points, we blend a similarity measurement that is sensitive to the large values’ locations in a vector with the adversarial loss. Both visual and numerical results of extensive experiments prove that the proposed model can obtain enhanced effectiveness and efficiency compared with some counterparts.

The main difficulty of synthetic aperture radar (SAR)-optical image matching or registration lies in the significant heterogeneous characteristics introduced by the different imaging mechanisms between SAR and optical images. Instead of directly using the raw image pair, transforming the pair into a feature domain, where they have homogeneous feature representation, is believed more effective. Inspired by image segmentation, we develop an end-to-end deep learning model for the SAR-optical matching, based on a siamese U-net with a fast Fourier transform (FFT) correlation layer. First, the siamese U-net with sharing weights extracts the feature maps of the SAR and optical images and projects the heterogeneous images into a homogeneous space. Then, the two feature maps are cross-correlated or normalized cross-correlated by the FFT layer and a similarity heatmap is obtained. Finally, the heatmap is send into a softmax2d classifier to determine the best matching, and thus matching is converted into classification. The nonlinear mapping capability of deep learning can well tackle the intensity variation across the different imaging modals; the encoder–decoder architecture with skip connections in the U-net can take full advantage of the global information and simultaneously preserve the local resolution and position information and thus guarantees high accuracy and robustness; besides, the FFT correlation is helpful for the efficiency improvement and training with large image pairs. Experiments show that the proposed method can achieve a pixel-level matching error.

In recent years, deep neural networks are increasingly popular in the field of radar high-resolution range profiles (HRRPs) target recognition. Unfortunately, recent researches have revealed that a deep-learning classifier can be easily fooled by adding small perturbations to the input, named adversarial attack. This provides us an inspiration for radar deceptive jamming signal generation in electronic countermeasures (ECMs). However, the perturbations generated by these adversarial attacks are usually of complex envelopes and quite low power, making it challenging for jammers to generate such actual jamming signals. To solve that issue, we propose a practical deceptive jamming generation method that learns the vulnerable range cells in an HRRP sample and injects several jamming pulses with specific amplitudes into these range cells. Such jamming signals are easy to generate and can deceive the radar automatic target recognition (RATR) model to output the wrong target category prediction with high confidence. To avoid the requirement of the recognition network structure information, we leverage the differential evolution optimization algorithm (non-gradientbased). Further, to provide the potential of real-time jamming signal generation during the test, an encoder is constructed not only to learn the separable features but also to find the vulnerable range cells and the specific amplitudes of the jamming pulses. In the experiments, we apply the proposed attack algorithms to fool the one-dimensional convolutional neural network-based HRRPRATR models. The extensive experimental results on measured aircraft HRRP dataset prove that the proposed algorithms achieve a promising attack performance and serve as a practical and fast deceptive jamming generation method.

Attitude determination of space target from an inverse synthetic aperture radar (ISAR) image sequence is an important but difficult task, because of the complex electromagnetic scattering characteristics of the target. To autonomously estimate the attitude of the space target from the ISAR image sequence, this letter introduces the key point feature extraction network (KPEN), and proposes a novel method for attitude determination via the key point feature extraction. The geometric relationship between the target attitude parameters and the key point feature is established through range-Doppler (RD) projection matrix. With the help of the key point extraction network, key point feature is extracted from ISAR image. Then the target attitude and its component size are autonomously recovered through key point feature by solving a nonlinear optimization. Extensive experiments show the effectiveness and superiority of the proposal.

We propose a meta network (MNet) for the problem of target-aspect missing in radar high-resolution range profile (HRRP)-based noncooperative target recognition, where a classifier must be generalized to new aspects not seen in the training set, given only a small number of HRRP data of each new aspect. The MNet is a time domain convolutional neural network (TCNN) that is built based upon recent progress in meta-learning. In effect, it learns a model that is easy and fast to fine-tune, allowing the adaptation to happen in the right space for fast learning. Besides, we construct a new controllable HRRP dataset suitable for the scenario of noncooperative target-aspect missing using electromagnetic simulation. Compared with the traditional methods, the MNet is more efficient and could achieve better performance. Extensive experiments on the simulated HRRP dataset are conducted to illustrate the effectiveness of the proposed method.