Suqin Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Suqin Wu has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Bakun Zhu
Department of Electronic and Optical Engineering, Space Engineering University, Beijing 101416, China
2 shared publications - Fusheng Wang
Department of Electronic and Optical Engineering, Space Engineering University, Beijing 101416, China
2 shared publications - Shuge Wang
Department of Electronic and Optical Engineering, Space Engineering University, Beijing 101416, China
2 shared publications - Wei Qu
Department of Electronic and Optical Engineering, Space Engineering University, Beijing 101416, China
2 shared publications Fusheng Wang
Department of Electronics and Optics, Space Engineering University, Beijing, China
1 shared publicationJiahao Wang
Department of Electronics and Optics, Space Engineering University, Beijing, China
1 shared publication
Latest Publications
- Open Access
- 16 November 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (22) , 3930
Physics-Driven Deep Learning for ISAR Imaging of Maneuvering Targets with a Large Rotation Angle
Inverse synthetic aperture radar (ISAR) is important for space target surveillance and recognition. However, under large rotation angles and uniformly accelerated target rotation, coupled high-order migration through range cells (MTRC) and two-dimensional space-variant phase errors cannot be accurately compensated by conventional methods based on small-angle or constant-rotation assumptions, causing image defocusing. To address this, a physics-driven subaperture differentiable back-projection (BP) ISAR imaging network is proposed. It embeds a parametric rotation model with angular velocity and acceleration: taking complex-valued HRRP sequences as input, the full aperture is partitioned into overlapping subapertures, and rotation parameters are regressed by a cross-subaperture parameter estimation network exploiting local scattering structures, temporal correlations, and motion evolution features; a differentiable BP imaging layer unifies rotation transformation, slant-range computation, complex-valued interpolation, phase compensation, and coherent accumulation into differentiable tensor operations, allowing image-domain losses to backpropagate to the parameter estimation module for closed-loop joint optimization of parameter estimation and focusing quality; and a phase registration and quality-guided coherent fusion mechanism adaptively suppresses sidelobes, defocusing, and noise from low-quality subapertures. Experiments on point-scatterer and electromagnetic simulation models demonstrate clear advantages, particularly at low SNR, providing a new path for high-resolution ISAR imaging of large-angle maneuvering targets.
- 16 November 2026
- Published by MDPI AG in Remote Sensing
- Vol. 18 (22) , 3930
Open AccessPhysics-Driven Deep Learning for ISAR Imaging of Maneuvering Targets with a Large Rotation Angle
Inverse synthetic aperture radar (ISAR) is important for space target surveillance and recognition. However, under large rotation angles and uniformly accelerated target rotation, coupled high-order migration through range cells (MTRC) and two-dimensional space-variant phase errors cannot be accurately compensated by conventional methods based on small-angle or constant-rotation assumptions, causing image defocusing. To address this, a physics-driven subaperture differentiable back-projection (BP) ISAR imaging network is proposed. It embeds a parametric rotation model with angular velocity and acceleration: taking complex-valued HRRP sequences as input, the full aperture is partitioned into overlapping subapertures, and rotation parameters are regressed by a cross-subaperture parameter estimation network exploiting local scattering structures, temporal correlations, and motion evolution features; a differentiable BP imaging layer unifies rotation transformation, slant-range computation, complex-valued interpolation, phase compensation, and coherent accumulation into differentiable tensor operations, allowing image-domain losses to backpropagate to the parameter estimation module for closed-loop joint optimization of parameter estimation and focusing quality; and a phase registration and quality-guided coherent fusion mechanism adaptively suppresses sidelobes, defocusing, and noise from low-quality subapertures. Experiments on point-scatterer and electromagnetic simulation models demonstrate clear advantages, particularly at low SNR, providing a new path for high-resolution ISAR imaging of large-angle maneuvering targets.
- Open Access
Azimuth-Frequency Conditioned Mamba U-Net for ISAR Image Refocusing
High-order phase defocus occurs in ISAR imaging of maneuvering targets. Traditional methods are limited by motion model mismatches, whereas existing deep imaging relies on complex-valued convolutions or self-attention mechanisms, whose excessive computational complexity hinders real-time processing. To address these issues, this paper proposes an Azimuth-Frequency Conditioned Mamba U-Net (AFC-Mamba U-Net) for ISAR image refocusing. The method treats the real and imaginary parts of the defocused complex image as dual-channel real-valued inputs, integrating linear-complexity state-space modeling with cross-layer feature fusion to significantly reduce computational and memory overhead while maintaining high refocusing quality. Specifically, we design an Azimuth-Frequency Conditional Mamba block (AFCMambaBlock), which performs bidirectional selective state-space scanning along the azimuth axis to capture long-range defocus dependencies with linear time complexity, and introduces azimuth-frequency conditional gating to incorporate frequency-domain phase error information into spatial features. A phase-state bottleneck is constructed to aggregate multi-scale enhanced states via a cross-scale state bank, forming globally phase-error-aware representations. Furthermore, we propose a state-guided skip connection mechanism that dynamically gates encoder skip features using bottleneck states, effectively suppressing the propagation of defocus artifacts while reducing redundant computations in the decoder. Experiments conducted on a simulated dataset containing eight types of spaceborne targets and real-world Yak-42 aircraft data demonstrate that the proposed method reduces MSE by 35.8% and NMSE by 41.5% compared to CNN-based U-Net, while achieving an SSIM of 0.9901. Compared to the state-of-the-art complex-domain network CVPHD, the proposed method achieves a significantly better efficiency–accuracy trade-off: it reduces model parameters and computational complexity by approximately two orders of magnitude, improves inference speed by more than fivefold, and cuts GPU memory usage by about 96%, while maintaining competitive structural consistency (SSIM > 0.99). Although CVPHD retains better MSE and NMSE on simulated data, the proposed method provides a lightweight and feasible solution for real-time and embedded ISAR refocusing.
Open AccessAzimuth-Frequency Conditioned Mamba U-Net for ISAR Image Refocusing
High-order phase defocus occurs in ISAR imaging of maneuvering targets. Traditional methods are limited by motion model mismatches, whereas existing deep imaging relies on complex-valued convolutions or self-attention mechanisms, whose excessive computational complexity hinders real-time processing. To address these issues, this paper proposes an Azimuth-Frequency Conditioned Mamba U-Net (AFC-Mamba U-Net) for ISAR image refocusing. The method treats the real and imaginary parts of the defocused complex image as dual-channel real-valued inputs, integrating linear-complexity state-space modeling with cross-layer feature fusion to significantly reduce computational and memory overhead while maintaining high refocusing quality. Specifically, we design an Azimuth-Frequency Conditional Mamba block (AFCMambaBlock), which performs bidirectional selective state-space scanning along the azimuth axis to capture long-range defocus dependencies with linear time complexity, and introduces azimuth-frequency conditional gating to incorporate frequency-domain phase error information into spatial features. A phase-state bottleneck is constructed to aggregate multi-scale enhanced states via a cross-scale state bank, forming globally phase-error-aware representations. Furthermore, we propose a state-guided skip connection mechanism that dynamically gates encoder skip features using bottleneck states, effectively suppressing the propagation of defocus artifacts while reducing redundant computations in the decoder. Experiments conducted on a simulated dataset containing eight types of spaceborne targets and real-world Yak-42 aircraft data demonstrate that the proposed method reduces MSE by 35.8% and NMSE by 41.5% compared to CNN-based U-Net, while achieving an SSIM of 0.9901. Compared to the state-of-the-art complex-domain network CVPHD, the proposed method achieves a significantly better efficiency–accuracy trade-off: it reduces model parameters and computational complexity by approximately two orders of magnitude, improves inference speed by more than fivefold, and cuts GPU memory usage by about 96%, while maintaining competitive structural consistency (SSIM > 0.99). Although CVPHD retains better MSE and NMSE on simulated data, the proposed method provides a lightweight and feasible solution for real-time and embedded ISAR refocusing.
- 10 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Antennas and Wireless Propagation Letters
- Vol. 24 (9) , 2959-2963
Physics-Informed Complex Natural Resonance Frequency Correction
Complex natural resonance frequency is only determined by target's intrinsic properties, and independent of incident angle and target's attitude, making target recognition based on complex natural resonance frequency a promising solution. However, the existing extraction methods inevitably introduce calculation errors, resulting in inaccurate extraction of it, which limits the improvement of recognition accuracy. To address these issues, Physics-Informed Complex Natural Resonance Frequency Correction (PI-CNRFC) is proposed. Firstly, a loss function based on target's resonance scattering mechanism is designed. Physics is used to drive network to extract features, strengthening the physical constraints on the corrected data. Secondly, by introducing the idea of Complex-valued Neural Network (CVNN), a Complex domain- Variational AutoEncoder (C-VAE) is designed, which considers the correlation between the real and imaginary parts of complex data,thereby excavating more internal features of data. Experimental results verify the effectiveness of the proposed method. Compared with the original data, the relative errors are reduced by 18.02% for the real part and 0.49% for the imaginary part. improving the consistency between corrected data and target's resonance scattering mechanism.
- 10 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Antennas and Wireless Propagation Letters
- Vol. 24 (9) , 2959-2963
Physics-Informed Complex Natural Resonance Frequency Correction
Complex natural resonance frequency is only determined by target's intrinsic properties, and independent of incident angle and target's attitude, making target recognition based on complex natural resonance frequency a promising solution. However, the existing extraction methods inevitably introduce calculation errors, resulting in inaccurate extraction of it, which limits the improvement of recognition accuracy. To address these issues, Physics-Informed Complex Natural Resonance Frequency Correction (PI-CNRFC) is proposed. Firstly, a loss function based on target's resonance scattering mechanism is designed. Physics is used to drive network to extract features, strengthening the physical constraints on the corrected data. Secondly, by introducing the idea of Complex-valued Neural Network (CVNN), a Complex domain- Variational AutoEncoder (C-VAE) is designed, which considers the correlation between the real and imaginary parts of complex data,thereby excavating more internal features of data. Experimental results verify the effectiveness of the proposed method. Compared with the original data, the relative errors are reduced by 18.02% for the real part and 0.49% for the imaginary part. improving the consistency between corrected data and target's resonance scattering mechanism.