Millimeter-wave radar is a promising alternative for human sensing due to its privacy preservation and robustness to illumination. However, the abstract nature of radar signals makes large-scale annotation expensive, causing data sparsity and label scarcity. While transfer learning approaches exist, they often rely on large labeled source datasets or discrete action categories. We propose FMSSL, a cross-modal self-supervised mmWave radar representation learning method via free-form motion. Unlike spectrogram-based or discrete methods, FMSSL aligns raw radar signals with synchronized skeleton trajectories to eliminate annotation dependencies. To resolve semantic ambiguity in free-motion data, we employ an Optimal Transport (OT)-based soft contrastive learning strategy for global distributional correspondence, combined with a rank-preserving constraint to maintain feature space topology. Additionally, a cross-modal masked trajectory modeling (MTM) task is introduced to capture generative temporal dynamics by reconstructing latent motion patterns in free-form motion. Experiments on downstream in-air handwriting and human action recognition demonstrate that FMSSL consistently outperforms contrastive and supervised baselines, exhibiting superior efficiency even with extremely limited labeled data.

During real flight missions, the terahertz synthetic aperture radar (THz-SAR) system often fails to follow a perfectly ideal path consistently. Nonideal trajectories induce parameter mismatches in conventional imaging methods, resulting in the invalidation of imaging algorithms and thus posing a significant challenge to high-precision imaging. To address this challenge, this research proposes an imaging method in the wavenumber domain for THz-SAR with variable-curvature trajectory (VCT). First, the rate of change of the radius of curvature (RCRC) is introduced to establish a variable-curvature model (VCM). Next, range resampling is employed to eliminate height errors in the radar data envelope, while a phase compensation function is constructed to remove height errors embedded in the phase, thereby equivalently projecting the 3-D spatial motion trajectory onto a 2-D planar curve. Subsequently, the RCRC is derived via the fitting of the 2-D planar curved trajectory, and the precise wavenumber spectrum is derived via the series inversion method, followed by the compensation of nonspace-invariant phase errors. Finally, phase functions are, respectively, constructed along the scene area in the wavenumber domain to compensate for residual phase errors, yielding high-precision imaging results. The effectiveness of the proposed algorithm for THz-SAR imaging under nonideal trajectory conditions is confirmed through both simulation and experimental results.

Millimeter-wave (mmWave) radar has attracted increasing research interest for contactless gesture recognition. However, in practical deployment, models trained on existing users often exhibit significant performance degradation when applied to unseen users. This decline arises because each user constitutes a distinct domain with unique behavioral patterns and physiological characteristics, leading to severe cross-domain distribution shifts in radar signal features. To address these challenges, we propose a two-stage few-shot gesture recognition adaptation method. During pretraining, a dual-prototype consistency strategy is proposed to explicitly align user-specific subdistributions within each gesture class, overcoming the limitation of conventional methods that treat multiple source users as a single monolithic domain and thereby enhancing the model's transferability to unseen users. In the adaptation stage, we perform source-free and lightweight fine-tuning using only a few labeled samples from the target domain while avoiding the computational burden and privacy concerns of joint training. Specifically, a variational autoencoder (VAE)-based latent space regularization with z-space alignment is employed to achieve stable and accurate recognition for the new target domain. Furthermore, knowledge distillation (KD) and L2 regularization toward the starting point (L2-SP) are incorporated to mitigate overfitting and prevent catastrophic forgetting of the source gesture knowledge. Experiments on a multisubject mmWave gesture dataset show that the proposed method significantly improves recognition performance for the target domains and maintains high accuracy on the original users. Visualization analyses further confirm effective latent space semantic alignment with minimal supervision.

Existingfew-shot object detection methods face a dual challenge when applied to few-shot synthetic aperture radar (SAR) target detection in cross-modal scenarios (optical -> SAR): First, high-dimensional noisy features and nonlinear modality differences make traditional feature-alignment mechanisms ineffective for matching optical and SAR modalities; second, current probabilistic classification frameworks, which rely on point estimates under maximum likelihood estimation, cannot adequately model the epistemic uncertainty induced by sample scarcity, leading to overconfident detection errors. To address these issues, we propose a projection-evidence collaborative optimization (PECO) method for cross-modal few-shot SAR target detection. Specifically, we first design a projection distribution alignment module, which constructs projected distributions and maps cross-modal data into a low-dimensional latent space, markedly reducing modality discrepancies and achieving effective cross-modal distribution alignment. Second, we introduce a dynamic uncertainty calibration module that models class probabilities with a Dirichlet evidence distribution and jointly optimizes epistemic and aleatoric uncertainties through a dynamic-weighting and label-driven calibration mechanism, thereby mitigating overconfidence errors in scarce-sample settings. Experimental results on the cross-modal datasets DIOR2SSDD and FAIR1M2SARAIRcraft verify the effectiveness of the proposed approach: PECO surpasses existing state-of-theart methods by 5.6% and 13.7%, respectively, in overall average detection performance, while also significantly improving model generalization.

Considering the practical limitations of power amplifiers, the transmit waveform with a low peak-to-average power ratio (PAPR) is necessary to prevent signal distortion. As a hot candidate waveform in integrated radar and communication applications, the orthogonal frequency division multiplexing (OFDM) signal exhibits high PAPR due to the random communication data it carries. To address this issue, we design a novel integrated signal by incorporating the Gaussian function and the proximal method of multipliers (PMM), while preserving the multi-subcarrier characteristics of the traditional OFDM signal. In this way, we not only retain the broadband characteristics but also achieve lower PAPR and better autocorrelation performance. Simulation results of synthetic aperture radar (SAR) imaging and communication are presented to demonstrate the performance of this waveform.

In complex and dynamic synthetic aperture radar (SAR) scenes, few-shot detection of novel classes suffers from sample scarcity and significant distribution differences between base and novel class features, leading to severe bias and poor generalization in existing few-shot object detection (FSOD) models. To address this issue, we propose a meta-transfer learning method based on dynamic semantic guidance (DSG). This approach combines the strengths of meta-learning and transfer learning, comprising three modules: semantic guidance (SG), distribution alignment metric (DAM), and global feature dynamic aggregation (GFDA). The SG module generates guided features with query semantic information to reduce the distribution gap between base and novel classes, dynamically adapting to few-shot novel class SAR targets. The DAM module applies adversarial training to achieve dynamic feature distribution alignment, improving model bias and generalization. The GFDA module dynamically aggregates and retains critical feature information, enhancing model detection performance. Experimental results on the SRSDD-v1.0, MSAR-1.0, and SAR-AIRcraft-1.0 datasets show that the DSG method outperforms state-of-the-art methods in the SAR field [Gaussian metafeature balanced aggregation (GMFBA)] and the optical domain [generalized FSOD(G-FSOD)], with average detection performance improvements of 1.21%, 1.45%, 1.44%, and 9.76%, 2.86%, 1.8%, respectively.

In the task of in-air handwriting recognition, millimeter-wave (mmWave) radar sensors offer advantages in low-power consumption, privacy protection, and robustness to environmental conditions. The traditional approach is to convert the radar echo signals into radar images by digital signal processing (DSP) algorithms and then perform trajectory tracking or recognition. In this article, we propose an innovative end-to-end recognition model that directly processes raw radar signals without designing specific DSP algorithms. In order to solve the problem of network training difficulties caused by high-dimensional raw radar signals, we introduce a multimodal feature alignment method based on variational analysis, which utilizes the common stroke pattern of handwritten trajectories to guide network training. Specifically, the method employs 2-D handwriting trajectory sequences to represent stroke patterns. Through the designed multimodal feature alignment algorithm, the raw signal features extracted by the end-to-end network gradually converge to the easily accessible 2-D handwriting trajectory features. Comparison experiments with traditional methods in complex handwriting recognition tasks demonstrate the superiority of the proposed method. Subsequent visualization analysis and ablation experiments further confirm the validity and interpretability of the model modules.

The imagery of high maneuverability synthetic aperture radar (SAR) is a challenging task due to the existence of the variable curvature trajectory. Firstly, in order to obtain sufficiently accurate slant range history, the rate of change in radius of curvature (RCRC) is introduced to establish an imaging geometry slant range model for high maneuverability synthetic aperture radar with constant RCRC and the equivalent four-order slant range polynomial is got based on this model. Then, we obtain the precise 2D spectra of the target with maneuvering radar platforms based on the principle of stationary phase and series reversion. Finally, the phase change introduced by the curvature change is compensated on the wavenumber domain, and the imaging process is carried out on this basis. Simulation results validate the feasibility of the proposed approach.

The terahertz (THz) synthetic aperture radar (SAR) system, known for its broad bandwidth and short synthetic aperture time, enables high-resolution imaging. However, due to the considerably smaller wavelengths at terahertz frequencies, multiple strong scattering centers often appear in imaging results, causing a loss of target details when using full aperture data directly. We propose a terahertz Synthetic Aperture Radar (SAR) adaptive aperture division back projection (BP) imaging algorithm based on target scattering characteristics. Through analyzing terahertz target scattering, we partition the aperture based on target scattering energy and calculate corresponding aperture weight coefficients. The final imaging result is obtained through a weighted coherent summation of sub-apertures. Experimental verification using a THz radar system with a carrier frequency of 0.3 THz and a bandwidth of 28.8 GHz confirms the effectiveness of the proposed algorithm.

Terahertz (THz) circular synthetic aperture radar (CSAR) demands higher precision in motion compensation (MOCO), posing significant challenges. To address this issue, this article proposes a multichannel MOCO algorithm based on adaptive subaperture division (MCAAD). First, the adaptive aperture division (AAD) algorithm is used to divide the image into multiple subapertures. Within each subaperture, initial coarse compensation is performed using data recorded by the inertial navigation system (GPS/INS). Subsequently, high-frequency vibration errors are compensated using the interferometric phase of different channels. Then, an image reconstruction algorithm is employed to image the compensated results, and a residual error estimation (REE) algorithm is applied to compensate for high-order phase errors and residual envelope errors, yielding well-focused subaperture images. Finally, the subaperture images are fused using a base-4 registration algorithm. The simulation and experimental results with measured data demonstrate that the proposed algorithm effectively compensates for THz CSAR imaging.

In this paper, an azimuth velocity estimation method based on spectral peak measurement-moving target shadow detection (SPM-SD) is proposed. Compared with the existing methods, this method takes into account the effect of the moving target position on the doppler shift and improves the accuracy of the azimuth velocity estimation value. The doppler modulation frequency of the azimuth phase is then estimated by using the fractional Fourier transform (FrFt), which in combination with the azimuth velocity gives an estimation of the distance velocity of the moving target. Finally, the estimation results are used to construct the first-order and second-order phase compensation functions for azimuth phase compensation, and the phase gradient autofocus algorithm (PGA) is used to compensate for the residual quadratic and higher phase errors. The real CSAR data processing results prove the effectiveness of the proposed method and greatly improve the imaging quality of the moving target.

Due to the high mobility and strong concealment characteristics of synthetic aperture radar (SAR) targets, the corresponding SAR datasets exhibit few-shot data properties, and there is a significant lack of research on few-shot target detection methods in the SAR domain. Furthermore, this study is subject to the following limitations: the scarcity of SAR data and significant sample variations make it difficult to control class centers using existing methods, and the learned models tend to be biased towards base classes while easily confusing novel classes with base classes. These limitations hinder the generalization of knowledge from base classes when detecting novel class targets. In this work, we propose a novel few-shot SAR target detection method based on Gaussian meta-feature balanced aggregation (GMFBA), which is based on meta-learning. Specifically, we first propose two novel feature aggregation methods with Gaussian metrics, namely Gaussian projection distribution metric (GPDM) and Gaussian kernel mean embedding metric (GKMEM). By estimating class distribution with variational autoencoders to replace traditional class prototypes, we sample from robust distributions and measure projection Wasserstein distance and Gaussian kernel mean embedding distance with prior distributions, obtaining the best robust support features under the optimal measurement results. Then, based on GPDM and GKMEM, we propose a novel balanced inter-class uncorrelated aggregation (BICUA) method, which extracts support features of each class according to the proportion of samples and aggregates them with query features in a balanced manner, promoting feature representation between different classes and ensuring no interference between features to significantly reduce confusion between base classes and novel classes. Specifically, GMFBA outperforms the state-of-the-art method G-FSOD significantly in all settings, achieving state-of-the-art performance. In contrast, the novel class detection performance of GMFBA has shown an average improvement of 8.56% on split1 and split2 of the SRSDD-v1.0 dataset, and an average improvement of 1.41% on split1 and split2 of the MSAR-1.0 dataset. The code is available at https://github.com/Caltech-Z/GMFBA.

High precision and high efficiency motion error compensation is a very challenging work for Terahertz synthetic aperture radar (THz-SAR) imaging, the tiny vibration of platform will make seriously THz-SAR image defocused. In this paper, a new compensation method based on the division of range sub-band conjugate, name as SBCM, is presented for THz-SAR effective motion correction. In the scheme, we use image entropy to iterate the overlap rate of the divided range sub-bands of THz-SAR echoes, and conjugate multiplying the division range-compressed results to obtain a more ideal imaging effectively. Both simulation and experiment results demonstrate the effectiveness of SBCM method. Compared with traditional echo-derived phase estimated autofocusing method, SBCM method not require parameter estimation and its arithmetic is small.

This paper introduces a method based on sub-aperture division for estimating the cross-track velocity of moving targets in terahertz circular synthetic aperture radar (CSAR). By carefully choosing the number of sub-apertures, the complete aperture is divided into multiple smaller apertures. In the sub-aperture regime, the small-angle approximation is applied to separate the cross-track velocity from other parameters related to moving targets. This approach allows for precise estimation of the cross-track velocity, facilitating accurate repositioning of the moving target. The effectiveness of the proposed method is confirmed through experimental results, evidencing its practicality.

Synthetic Aperture Radar (SAR) targets often exhibit characteristics such as high mobility and strong concealment, resulting in scarce SAR data and the manifestation of few-shot data properties. These few-shot SAR targets are susceptible to interference from complex background information and mutual interference of target features, making it challenging to distinguish SAR targets from the background. Additionally, there is confusion in features among different targets, leading to models being highly insensitive to few-shot SAR targets under complex distribution conditions in new tasks. Similarly, these few-shot SAR targets exhibit significant sample scarcity and sample variations, resulting in pronounced fluctuations in class centers and difficulty in determining sample distributions. This leads to challenges in accurately representing the potential representative features of few-shot SAR targets by the model. To address these issues, further enhancement of SAR target features is necessary to provide a robust foundation for the ultimate aggregation module. Therefore, based on the meta-learning paradigm, we propose a method for few-shot target detection in SAR imagery via intensive meta-feature aggregation (IMFA), aiming to reinforce SAR target features for improved representation. Specifically, firstly, we propose a novel hierarchical multi-head cross attention (HMCA) to capture global multiscale contextual information in different subspaces and analyze representative features between different targets to distinguish SAR targets from the background. Then, based on HMCA, we introduce a novel feature coupling module (FCM) to couple support features with cognitive information from the query image on the support branch. This is done to reduce the confusion and mutual interference of features between targets while enhancing the model’s generalization ability on new tasks. Finally, on the support branch with query-aware information, we construct a Gaussian distribution to estimate the class distribution of few-shot SAR targets and replace traditional class prototypes. On this basis, we propose the feature information maximization module (FIMM) to avoid feature information shift, greatly strengthening the expression of potential features. Through these steps, reinforced meta-features can be obtained, enabling efficient aggregation. Experiments on the SRSDD-v1.0 and MSAR-1.0 datasets demonstrate that our method has consistently outperformed state-of-the-art approaches in all configurations, achieving state-of-the-art performance.

This letter addresses the problem of changes in target shape and structure during imaging caused by different height structures being focused at the same height in the terahertz (THz) band. A new circular synthetic aperture radar (SAR) imaging algorithm based on the extraction of target structure scattering characteristics is proposed to solve this phenomenon. First, a circular SAR (CSAR) imaging model is established for different heights. The target was imaged at different heights, resulting in a series of image sequences in different heights and azimuth directions. Then, incoherent tracking of the backscattered energy is used to search for effective scattering centers at different heights. The scattering characteristic curves of different structures are obtained based on the effective scattering centers. Finally, the effectiveness of the proposed algorithm is verified by simulation and measured data. The proposed algorithm has accurately focused the target, while also enhancing the details. In addition, the contrast-to-noise ratio (CNR) of the imaging results has been improved by 4.87% compared to traditional algorithms.

High-value novel-class SAR targets exhibit distributional differences from base-class SAR targets, leading existing FSOD methods to suffer from model bias and poor generalization. To address this issue and more effectively detect few-shot targets, this paper introduces a semantic compensation method based on meta-learning (SCML). Specifically, we generate support compensation features enriched with query semantic information to mitigate distribution discrepancies and guide the model in adapting to new-class SAR targets under few-shot conditions, thereby significantly improving detection performance. On existing public datasets, our method demonstrates superior detection accuracy compared to other SOTA approaches.

In the past, synthetic aperture radar (SAR) target detection research has mainly focused on the representation of bounding boxes. However, the distant contextual prior knowledge in SAR images has not been fully utilized, leading to challenges in detecting small targets and densely arranged targets with multi-scale characteristics. Therefore, this paper proposes a high-precision rotating target detector (HRTD) for SAR scenarios to address the aforementioned issues. More precisely, we introduce a multi-scale rotation-equivariant network (MRNet) and align rotation-invariant regions of interest to enhance target orientation prediction capability. The multi-scale dynamic global attention (MDGA) in MRNet can dynamically adjust the contextual multi-scale spatial receptive field in real-time, facilitating the comprehensive extraction of distant contextual features. This method achieves significantly better detection performance on the SSDD+ and SRSDD-v1.0 datasets compared to other algorithms.

The Traditional Range-Doppler algorithm is completed by two-dimensional matched filtering, so there are often sidelobes which may affect synthetic aperture radar (SAR) imaging performance. However, recent studies have shown that CP-OFDM SAR imaging algorithm, which using orthogonal frequency-division multiplexing (OFDM) signals with sufficient cyclic prefix (CP), can obtain ideally zero sidelobes and build an inter-range-cell interference (IRCI)-free SAR image, but with a constraint that the module of each subcarrier’s coefficient should be constant and nonzero. Considering some practical applications, the subcarriers within a certain bandwidth may not necessarily be continuously allocated, thus "0" may appear in the frequency domain vector of an OFDM signal, making it difficult to complete range compression. In this paper, we combine the existing CP-OFDM SAR imaging algorithm with generative adversarial imputation nets (GAIN) to solve this problem, and simulation results are presented to illustrate the performance of this method.

As a widely used communication signal, orthogonal frequency division multiplexing (OFDM) signal has good sensing performance. It is the key technology of the integration of sensing and communication (ISAC), which is the potential core technology of the sixth generation mobile communication. The fusion of Multiple input and multiple output (MIMO) technology and OFDM signal was conducted to improve the communication speed and imaging resolution of ISAC system. In our work, the multi-signal classification (MUSIC) algorithm is adopted to process MIMO-OFDM data and achieve time of arrival (TOA) and direction of arrival (DOA) estimation. Based on the simulations and experiments in a variety of multi-object scenes, it shows that the MUSIC algorithm can restore the position of the objects more accurately and more robustness than traditional ways.