Three-dimensional shape retrieval from synthetic aperture radar (SAR) images has long presented a significant research challenge, with single-view reconstruction proving even more complex due to constraints such as the scarcity of labeled data, limited sample diversity, and heightened sensitivity to radar scattering characteristics. Recently developed deep learning-based methods have made progress in single-view target reconstruction from SAR images. However, these methods still rely heavily on 3-D ground-truth supervision and fail to fully leverage SAR imaging mechanisms for 3-D reconstruction. To address these limitations, an end-to-end unsupervised single-view 3-D reconstruction framework based on a differentiable SAR renderer (DSR) is proposed, achieving precise reconstruction while eliminating the need for ground-truth data. Specifically, the encoder-decoder architecture effectively extracts 3-D and angular features, utilizing template deformation to preserve both fine details and global structures across scales, along with essential pose information for 3-D shape reconstruction. The reconstructed 3-D model is projected onto a 2-D plane, and pixel-level intersection over union (PIoU) loss is employed for unsupervised learning, enabling the extraction of discriminative latent structures and patterns. This approach effectively reduces low-frequency noise, sharpens critical edges, and enhances high-frequency details, improving spatial structure accuracy while minimizing shape distortions and height errors in complex targets. Extensive quantitative and qualitative experiments on both simulated and real datasets demonstrate the framework’s superior performance in single-view SAR 3-D target reconstruction, offering a promising solution with broad potential applications.

The electromagnetic inverse task has long been recognized as a challenging research problem, which attracted substantial attention from the microwave community. In this paper, our objective is to explore the intricate relationship between geometric model imaging and radar view angles in Synthetic Aperture Radar (SAR) images, mainly focusing on the inverse problem of radar view angle estimation given a target model. However, the high cost and limited availability of SAR data acquisition, along with background interference and complex imaging mechanisms in SAR images, present significant challenges to the generalization, robustness, and accurate feature extraction of existing methods. To address these issues, we propose an interactive deep reinforcement learning (DRL) framework, which facilitates the interaction between the agent and an embedded electromagnetic simulator environment to simulate a human-like process of angle prediction step-by-step. A large number of experimental results verified that the proposed method can accurately predict the radar perspective of SAR images.

This article attempts to infer the orientation angle of the target in synthetic aperture radar (SAR) images using reinforcement learning (RL). It is intended to address the challenges like limited interpretability, the scarcity of SAR data, and complex imaging mechanisms restrict the broader application of learning-based approaches. We propose an interactive deep RL (DRL) framework, where an electromagnetic simulator named differentiable SAR renderer (DSR) is embedded to facilitate the interaction between the agent and the environment. Specifically, DSR generates SAR images at arbitrary orientation angles in real time, helping to simulate a human-like process of angle estimation. The differences in sequential and semantic aspects between images of different orientation angles are leveraged to construct the state space in DRL, which effectively suppress the complex background interference, and enhance the sensitivity to temporal variations. Moreover, to maintain the stability and convergence of our approach, reward mechanisms such as memory difference, smoothing and boundary penalty are incorporated to contribute to the formulation of the comprehensive reward function. Extensive experiments performed on both simulated and real datasets demonstrate the effectiveness and robustness of our proposed method. In addition, when utilized in the cross-domain area, the proposed method mitigates inconsistency between simulated and real domains, outperforming reference methods significantly.

Synthetic Aperture Radar (SAR) is extensively employed in both civilian and military sectors, with recent advancements leveraging deep learning for automatic SAR image interpretation. However, the effectiveness of these techniques, particularly Convolutional Neural Networks (CNN), is challenged by insufficient angle range in actual samples due to satellite incident angle constraints. This article proposes a method for generating multi-view samples of SAR targets based on a CNN module integrated with Differentiable SAR Renderer (DSR). Specifically, a polygon mesh is reconstructed from two-dimensional (2D) SAR images through the CNN module, and the DSR is utilized to reversely render SAR target images of various viewpoints from the reconstructed mesh, including the samples used to match with original input 2D images. Then, the generated images is used to compute the loss in training phase, and no three-dimensional (3D) ground truth is required. Experiments are conducted on simulated SAR images and the results demonstrate the efficacy of multi-view sample generation for SAR targets.

Three-dimensional (3D) reconstruction of complex targets on the ground from multi-view synthetic aperture radar (SAR) images is of great interests. The inherently-integrated forward-inverse architecture of the differentiable SAR renderer (DSR) provides a promising solution to the general inverse problem of SAR target reconstruction. In this context, the target’s shadow provides complementary information to its scattering image. Hence, this paper proposes a novel DSR-based target reconstruction approach using both the target image and its shadows. The capabilities of DSR are extended to generate not only target scattering images but also shadows. Furthermore, the gradients of the outputs, specifically illumination map and shadow map, with respect to the inputs, i.e., target geometry represented as a mesh, are derived. This enables us to develop a gradient-descent inverse approach for solving the general reconstruction problem. Extensive simulations and quantitative evaluations demonstrate that incorporating both the target scattering image and its shadows significantly improves the reconstruction performance. Moreover, our analyses indicate that achieving optimal reconstruction effects requires a minimum of 9 views with a relatively even distribution. Finally, the proposed algorithm is validated using real SAR images of vehicle targets.

Forward modeling of wave scattering and radar imaging mechanisms is the key to information extraction from synthetic aperture radar (SAR) images. Like inverse graphics in the optical domain, an inherently-integrated forward-inverse approach would be promising for SAR advanced information retrieval and target reconstruction. This paper presents such an attempt at inverse graphics for SAR imagery. A differentiable SAR renderer (DSR) is developed, which reformulates the mapping and projection algorithm of the SAR imaging mechanism in the differentiable form of probability maps. First-order gradients of the proposed DSR are then analytically derived, which can be back-propagated from rendered image/silhouette to the target geometry and scattering attributes. A 3D inverse target reconstruction algorithm from SAR images is devised. Several simulation and reconstruction experiments are conducted, including targets with and without background, using synthesized data or real measured inverse SAR (ISAR) data by ground radar. Results demonstrate the efficacy of the proposed DSR and its inverse approach.