Search

With the rapid development of remote sensing (RS) technology, high-resolution RS image change detection (CD) has been widely used in many applications. Pixel-based CD techniques are maneuverable and widely used, but vulnerable to noise interference. Object-based CD techniques can effectively utilize the abundant spectrum, texture, shape, and spatial information but easy-to-ignore details of RS images. How to combine the advantages of pixel-based methods and object-based methods remains a challenging problem. Besides, although supervised methods have the capability to learn from data, the true labels representing changed information of RS images are often hard to obtain. To address these issues, this article proposes a novel semisupervised CD framework for high-resolution RS images, which employs small amounts of true labeled data and a lot of unlabeled data to train the CD network. A bihierarchical feature aggregation and extraction network (BFAEN) is designed to achieve the pixelwise together with objectwise feature concatenation feature representation for the comprehensive utilization of the two-level features. In order to alleviate the coarseness and insufficiency of labeled samples, a confident learning algorithm is used to eliminate noisy labels and a novel loss function is designed for training the model using true- and pseudo-labels in a semisupervised fashion. Experimental results on real datasets demonstrate the effectiveness and superiority of the proposed method.

This paper is focused on the problem of resolving the range ambiguity in a fixed phase coding-phased array (FPC-PA) radar. At the modeling stage, the FPC-PA radar is configured by a fixed phase coding in both transmit elements and pulses, where the transmit beam is modulated with a coding phase in each pulse. Then, after some appropriate signal processing procedures, the degrees-of-freedom (DoFs) in the transmit domain is obtained, and the range region numbers, namely the index of the range ambiguity region, are discriminated in the transmit spatial frequency domain. Particularly, echoes from different range ambiguity regions correspond to different pointing directions of the equivalent transmit beampattern. Moreover, in order to obtain the range ambiguity index and angle of the target, a three-stage algorithm is developed to enhance the performance against the effect of amplitude and phase errors. At the analysis stage, the parameter estimation performance is assessed in comparison with the Cramér-Rao lower bound (CRLB). Experimental and numerical results are provided to demonstrate the effectiveness of resolving range ambiguity and parameter estimation with the proposed method.

This article is focused on resolving the problem of suppressing mainbeam deceptive jammers in the element-pulse coding (EPC) radar. At the modeling stage, the element-pulse phase coding matrix (EPPCM) is employed at the transmit side, introducing extra coding degrees-of-freedom (DOFs) for the discrimination of the true target and mainbeam jammers. By utilizing the extra coding DOFs, nonadaptive and adaptive suppression method for mainbeam jammers are developed. Firstly, in order to suppress the jammers non-adaptively, the EPPCM is designed based on the coding steering vector orthogonalization (CSVO) method, which requires the identical angle of the true and false targets. Then, the adaptive suppression method is proposed for the general cases, where the angle deviation between the true and false targets occurs. To this end, the unitary EPPCM is designed to achieve the coding DOFs at the transmit side. Moreover, we take the towed radar active decoy (TRAD) jammer as an example and the sample selection scheme for the estimation of jammer-plus-noise covariance matrix is proposed. At the analysis stage, experimental and numerical results in terms of output signal-to-jammer-plus-noise ratio (SJNR) demonstrate the effectiveness of the devised methods for suppressing mainbeam deceptive jammers.

Ground moving target detection is one of the foremost tasks for multichannel synthetic aperture radar (SAR) system. The traditional robust principal component analysis (RPCA) method is capable of separating low-rank and sparse components from mixed echo signals, and it has been widely applied in SAR ground moving target indication (GMTI). However, it suffers from sensitivity to channel mismatch, high computational complexity, and excessively high false alarm rates. To address these issues, a novel method that combines adaptive multichannel data reconstruction (DR) with improved pseudo-skeleton decomposition (IPSD) is proposed. First, the iterative weighted approach is presented to precisely reconstruct the multichannel data vector with the joint-pixel model. After that, IPSD is presented to achieve the moving target detection, in which the row and column index sets are selected using the generalized inner product (GIP) and the amplitude histogram distribution criterion. Compared to the existing algorithms, the proposed algorithm effectively addresses the challenge of improving local region coherence in multichannel image sequences. In addition, compared to previous RPCA methods, the proposed algorithm significantly reduces false alarm rates in strong clutter backgrounds while achieving higher efficiency. Simulation results and real SAR data experiments validate the effectiveness of the proposed algorithm.

This article aims to investigate intelligent strategies of interference suppression for radar systems in the background of complex electromagnetic interferences. At the modeling stage, an interactive loop is established exploiting the interaction between the radar and the environment for interference suppression based on reinforcement learning. Specifically, the mappings from the interference suppression to the reinforcement learning, including the interference state set, the method set, evaluation criteria of interference suppression in different domains, and the principle of interference substate transformation, have been established. In this respect, two algorithms, including the Retroactive-Q (R-Q) learning and Retroactive-Deep Q Network (R-DQN), are developed by introducing a backtracking Q-value, which links the evaluations in each time step of a training round. At the analysis stage, the selection probabilities of the optimal implementation sequence for interference suppression are studied, and comparisons among the devised R-Q learning, R-DQN, conventional Q learning, and DQN are carried out in terms of output Q-values. Numerical results corroborate the effectiveness and robustness of the considered suppression strategies in diverse scenarios.

In this article, mainlobe deceptive jammer mitigation for multiple targets is investigated using a subaperture-frequency diverse array-multiple-input multiple-output (MIMO) radar. At the modeling stage, the transmit array is divided into multiple overlapped subapertures consisting of fully overlapped secondary subarrays, where a small frequency increment is introduced among different subapertures, while a phase coding modulation is adopted among different secondary subarrays. In this respect, all targets and jammers can be separated in a 3-D space-time cube, including the transmit spatial domain, the slow-time pulse domain, and the receive spatial domain. In this case, the problem of jammer mitigation is formulated, and a two-stage beamforming method is proposed, where the jammers with delayed pulses are first suppressed, and the jammers, which are located within the same transmit pulse, are further mitigated. Specifically, an alternating procedure is developed by transforming the suppression problem into several 1-D subproblems. In order to further improve the jammer mitigation performance, the subaperture transmit weights are also designed by maximizing the worst-case output signal-to-jammer-plus-noise ratio (SJNR) resorting to an iterative successive convex approximation (ISCA) algorithm. Numerical results corroborate the effectiveness of the developed methods in mitigating mainlobe deceptive jammers for the case of multiple true targets.

Separability is an important criterion for measuring the degree of overlap between different features in pattern recognition. Stronger separability indicates clearer feature margins, facilitating classification of samples into their respective classes. Therefore, if the separability of radar target echoes within the feature space increases, the likelihood of correct classification increases. A waveform with enhanced separability offers more reliable discrimination for specific targets and highlights their unique aspects. Accordingly, this paper proposes radar waveform design methods driven by target echo pattern separability, i.e., the convolution of the target high-resolution range profiles (HRRP)/target impulse response (TIR) and transmit waveform. First, from the views of local HRRP/TIR manifold structure preservation and inter-class distance enlargement, we construct a minmax based fractional optimization model with non-convex and non-linear orthonormality and constant modulus (CM) constraints. Then, the derived waveform design solution is computed iteratively via subproblem division, fraction simplification, and high-order polynomial optimization. Furthermore, we extend our study to a linear discriminant analysis-based waveform design model, which incorporates minmax intra-class distance and inter-class distance metrics to maximize separability of all classes from a worst-case perspective. We utilize simulated datasets based on the scattering point model as well as electromagnetic simulation and MSTAR CSV datasets to evaluate the performance of our waveform design approaches.