13 March 2023 in IEEE Transactions on Geoscience and Remote Sensing
Micro-Doppler analysis of rigid-body target is significant for attitude estimation and recognition of space objects. The traditional micro-Doppler analysis method for rigid-body targets includes two sequential steps. First, the time-frequency analysis is performed on the radar echo data of the target, and then the micro-Doppler curve of each scattering center is separated and extracted from the time-frequency map. The second step depends on the micro motion model of the target. In the micro-Doppler analysis of real rigid-body target, there are some problems such as the mismatch of the micro motion model, the incidence angle dependence of the scattering center position, and the partial occlusion of the scattering center. Therefore, it is very difficult to correctly extract the micro-Doppler curves of multiple scatterers. In this paper, an end-to-end micro-Doppler analysis method for rigid-body target based on deep learning network is proposed, which can directly separate and extract the micro-Doppler curves of multiple scatterers from the target echo data. Specifically, an Asymmetrical AutoEncoding (A2E) network equipped with a Data Pre-Processing (DP2) module is developed to extract Time-Frequency Curves (TFCs) from radar echos. Considering the sparseness of Time-Frequency Distribution (TFDs), we then develop a novel Energy-Concentration Objective (ECO) function based on Min-Max game to enhance curves energy while suppress the background energy. In practice, measured TFDs are rarely annotated, it restricts the generalization capability of the A2E from the simulation to the measurement. To bridge the gap, two-fold modifications are finally constructed: i) we insert a partial-shared branch of decoder to reconstruct the TFD from the DP2 module; ii) we regularize the ECO function with a knowledge preservation based reconstruction bound to further capture the characteristics of measured echos in a semi-supervised way at test time to relieve the domain-shift problem.
11 November 2024 in IEEE Transactions on Aerospace and Electronic Systems
Deep Neural Network (DNN)-based Synthetic Aperture Radar (SAR) automatic target recognition (ATR) methods have been shown to be vulnerable, particularly in low-data scenarios, also known as generalized few-shot learning situations. This vulnerability significantly restricts their real-world applications. Therefore, enhancing the generalization capability of DNN-based methods is crucial for practical SAR target recognition. To address this challenge, we observe that misclassified samples in generalized few-shot SAR target recognition often exhibit similar behaviors to adversarial samples-they tend to be located near the classification boundary and are distant from their respective class centers. To tackle this issue, we propose a Max-Mahalanobis Centers (MMC) guided Adversarial Network (M2cgAN). This approach focuses on learning compact representations that cluster around preset optimal centers for different targets, thereby reducing the likelihood of misclassifying SAR images. Specifically, we first utilize Max-Mahalanobis Centers (MMC) due to its property that if the input is distributed according to a Max-Mahalanobis distribution (MMD), a linear classifier will achieve optimal robustness against misclassified samples. Building on this, we leverage this property in generalized few-shot SAR Automatic Target Recognition (ATR) by employing a Deep Neural Network (DNN), specifically ResNet12, to project SAR images onto the preset optimal centers defined by MMC. Finally, an Adaptive Area Perception (A2P) module has been developed to extract the most discriminative regions in SAR images, effectively reducing the impact of strong scattering points in complex backgrounds. The effectiveness and efficiency of M2cgAN are validated through comprehensive experiments on the widely used MSTAR and OpenSARShip benchmarks.