Xingzhao Liu

Ground Penetrating Radar (GPR) radargram are affected by antenna coupling reflections and interference from anisotropic soil media, introducing significant clutter that severely hinders subsequent target inversion and detection. Supervised learning methods rely on paired cluttered and target image, acquiring such data in real experimental scenarios is often impractical. To overcome this challenge, we design SADGPR-Net, a general unsupervised self-augmented dual-branch network designed to eliminate clutter in unpaired GPR images. We design a dual-branch parameter-sharing Decoupler–Synthesizer mechanism, which guiding the Decoupler to improve denoising performance. In the dual-branch design, Branch 2 employs the Synthesizer to generate paired data, enhancing Decoupler performance via consistency constraints and mitigating degradation from unpaired data in unsupervised networks. Branch 1 shares parameters with Branch 2 and is jointly optimized, collaboratively enhancing the Decoupler’s effectiveness. Additionally, we introduce Wasserstein loss to stabilize the training process and perceptual loss to enhance consistency. In this study, we validate the framework primarily on pipeline scenarios. Experimental results demonstrate that, compared to CycleGAN, our method improves PSNR by 7.11 dB and SSIM by 0.0148, while reducing MAE, LPIPS, DISTS, and FID by 0.45, 0.0228, 0.0691, and 34.24, respectively. These results demonstrate the superiority of our proposed method.

Target angle estimation with a single-channel wide beam antenna is a fundamental and crucial task in radar and remote sensing. Traditional target-angle estimation methods primarily rely on the phase differences of electromagnetic waves between array antennas, resulting in a high level of complexity in the antenna system. Furthermore, the angle estimation accuracy is also limited by the array antenna aperture. To alleviate the problems of system complexity and aperture limitations, a multi-media nonlinear wavefront modulation system (MN-WMS) for angle estimation is proposed. This system does not require modification of the original wide-beam antenna; therefore, the system complexity is low. By introducing nonlinear wavefront modulation information, the target angle can be accurately estimated. Specifically, we first propose a wavefront modulation angle estimation principle that introduces additional reference information for the angle estimation task. Then, we propose a multi-media modulator for electromagnetic waves modulation and present the implementation of the modulator. Finally, a multi-media nonlinear wavefront modulation angle estimation algorithm (MN-WMA) is proposed, which can be adaptively adjusted to balance processing time and accuracy require ments. High-accuracy angle estimation for a single-channel wide beam antenna can be effectively realized. Microwave anechoic chamber experiments show that the method achieves an angle-measurement accuracy of 0.163◦ with the single-channel wide beam antenna, verifying the effectiveness of the proposed method. This is a new exploration of using the nonlinear wavefront modulation method for angle estimation with a single-channel wide-beam antenna.

Semi-supervised change detection (SSCD) has emerged as a promising approach to reduce the burden of pixel-level annotation in bi-temporal remote sensing imagery (RSI). However, learning robust and reliable representations under style variations remains a key challenge, as appearance discrepancies across temporal acquisitions often account for a large portion of the visual differences between bi-temporal images. These challenges can be attributed to two levels of style–semantic entanglement: (1) cross-temporal style discrepancies caused by acquisition variations (e.g., illumination shifts and seasonal changes), which may confuse the model and increase sensitivity to superficial appearance differences; and (2) intra-temporal style heterogeneity introduced by diverse background characteristics (e.g., land-cover diversity and vegetation density differences), which can bias feature learning and obscure meaningful semantic changes. To address these challenges, we propose StyleDe-CD, a dual-level style disentanglement framework for SSCD. At the input level, we introduce stochastic spectral–photometric style augmentation (SSPSA), which integrates spectral style alignment (SSA) and photometric transformations (PT) through a stochastic switching mechanism. SSA alleviates cross-temporal style discrepancies by smoothly blending the low-frequency amplitude spectra of bi-temporal images using Gaussian-weighted soft masks, while PT introduces diverse appearance perturbations to improve robustness to style variations. At the feature level, we develop intra-temporal style mixing (ISM), which stochastically mixes channel-wise feature statistics across corresponding temporal components of shuffled image pairs within the same minibatch, encouraging the model to reduce sensitivity to instance-specific background styles and to emphasize semantic change cues. Extensive experiments on the LEVIR-CD, WHU-CD, and CDD benchmarks demonstrate that StyleDe-CD consistently achieves strong and competitive performance compared to existing semi-supervised baselines, particularly under low-label regimes, indicating its effectiveness in SSCD.

Camera pose estimation from optical images is well established, enabling autonomous navigation and mapping. In contrast, synthetic aperture radar (SAR) pose estimation remains underexplored and immature, with existing efforts typically relying on point features extracted from single-channel SAR images. To fully exploit the geometric cues in SAR imagery, we propose a novel pose estimation method that leverages point and line features extracted from multi-channel interferometric SAR data. The approach operates in two stages. At the radar level, a Circular-n-Point-and-Line problem is formulated using point and line features extracted from SAR images and a known 3D scene. A unified objective function combining point and line reprojection errors is constructed, and the radar-level pose is estimated using a Lie-algebra-based conjugate gradient method. At the array level, we formulate a matrix equation that relates inter-channel displacements of scatterers to the unknown array parameters. This equation is solved using Sequential Quadratic Programming. Experimental results demonstrate that the proposed approach achieves accurate and robust pose estimation at both radar and array levels, offering a viable solution for navigation in environments where external positioning systems are unavailable.

Overcoming the angular resolution limitations imposed by antenna apertures remains a critical challenge in radar systems. The nonlinear wavefront modulation technique enables angular resolution beyond the limitations of antenna beamwidths while maintaining the physical aperture size. This article proposes a novel method to enhance angular resolution within a single beamwidth by employing nonlinear wavefront modulation and investigates both the theoretical and experimental aspects of this technology. First, a wavefront modulation signal model is established, and the resolution enhancement factor (REF) is defined to quantify the performance of angular resolution enhancement. Two analytical criteria—the subspace distance criterion and Cramér–Rao bound (CRB) criterion—are proposed to evaluate resolution performance. An $L$ -band verification system is constructed, and experiments are conducted in a microwave anechoic chamber. Results validate the angular resolution enhancement principle and confirm the system’s ability to resolve multiple targets within the single beamwidth in both azimuth and elevation. In the experiment, our method successfully discriminates multiple targets with a minimum azimuth separation of 1.08° and a minimum elevation separation of 5.14°, despite the broad beamwidths of the receiving antenna (67° in azimuth and 98° in elevation). Finally, a comprehensive analysis assesses the system under varying signal-to-noise ratios (SNRs), numbers of wavefront modulation states, and calibration errors. This study provides a rigorous foundation for the principles and engineering implementation of nonlinear wavefront modulation, offering significant utility for future radar system design.

Despite notable advancements in remote sensing vision-language models (VLMs), existing models often struggle with spatial understanding, limiting their effectiveness in real-world applications. To push the boundaries of VLMs in remote sensing, we specifically address vehicle imagery captured by drones and introduce a spatially-aware dataset AirSpatial, which comprises over 206K instructions and introduces two novel tasks: Spatial Grounding and Spatial Question Answering. It is also the first remote sensing grounding dataset to provide 3DBB. To effectively leverage existing image understanding of VLMs to spatial domains, we adopt a two-stage training strategy comprising Image Understanding Pre-training and Spatial Understanding Fine-tuning. Utilizing this trained spatially-aware VLM, we develop an aerial agent, AirSpatialBot, which is capable of fine-grained vehicle attribute recognition and retrieval. By dynamically integrating task planning, image understanding, spatial understanding, and task execution capabilities, AirSpatialBot adapts to diverse query requirements. Experimental results validate the effectiveness of our approach, revealing the spatial limitations of existing VLMs while providing valuable insights. The model, code, and datasets will be released at https://github.com/VisionXLab/AirSpatialBot

As for the clutter suppression based on dimension-reduced space-time adaptive processing (STAP) in a spaceborne multichannel radar system, the performances of different auxiliary channels are theoretically analyzed in this paper. Then an improved dimension-reduced STAP based on hybrid auxiliary channels is proposed. In this method, the localized angle-Doppler domain processing is chosen and a hybrid structure of auxiliary channels from sliding windows and adjacent channels is adopted to suppress the clutter spectral leakage components. The selection method of auxiliary channels is also analyzed. Compared with other STAP methods, the proposed method can fully suppress the clutter spectral leakage components and approach an ideal target detection performance, while requiring low computation complexity and a small number of independent and identically distributed samples.

Ground-penetrating radar (GPR) is an important geophysical technique in subsurface detection. However, traditional numerical simulation methods such as finite-difference time-domain (FDTD) face challenges in accurately simulating complex heterogeneous mediums in real-world scenarios due to the difficulty of obtaining precise medium distribution information and high computational costs. Meanwhile, deep learning methods require excessive prior information, which limits their application. To address these issues, this paper proposes a novel two-step forward modeling strategy for GPR data of metal pipes. The first step employs the proposed Polarization Self-Attention Image Translation network (PSA-ITnet) for image translation, which is inspired by the process where a neural network model “understands” image content and “rewrites” it according to specified rules. It converts scene layout images (cross-sectional schematics depicting geometric details such as the size and spatial distribution of underground buried metal pipes and their surrounding medium) into simulated clutter-free GPR B-scan images. By integrating the polarized self-attention (PSA) mechanism into the Unet generator, PSA-ITnet can capture long-range dependencies, enhancing its understanding of the longitudinal time-delay property in GPR B-scan images. which is crucial for accurately generating hyperbolic signatures of metal pipes in simulated data. The second step uses the Polarization Self-Attention Style Transfer network (PSA-STnet) for style transfer, which transforms the simulated clutter-free images into data matching the distribution and characteristics of a real-world underground heterogeneous medium under unsupervised conditions while retaining target information. This step bridges the gap between ideal simulations and actual GPR data. Simulation experiments confirm that PSA-ITnet outperforms traditional methods in image translation, and PSA-STnet shows superiority in style transfer. Real-world experiments in a complex bridge support structure scenario further verify the method’s practicability and robustness. Compared to FDTD, the proposed strategy is capable of generating GPR data matching real-world subsurface heterogeneous medium distributions from scene layout models, significantly reducing time costs and providing an efficient solution for GPR data simulation and analysis.