多地面运动目标大动态SAR成像稀疏表示

Sparse representation of large dynamic range SAR imaging for multiple ground moving targets

为了保证对多个地面运动目标同时进行合成孔径雷达成像时具有足够的响应动态范围,提出了一种基于参数化贝叶斯机器学习的压缩感知稀疏表示方法,在对运动目标稀疏特征增强的同时可以显著地提高多目标合成孔径雷达成像的响应动态范围。首先,利用渐进线性的吕氏分布时频表示方法获得多运动目标的多普勒调制参数,并构建二阶多项式傅里叶字典;然后,针对该字典可能导致的压缩感知有限等距特性欠佳的问题,研究利用字典的互相关度进行定量评估;最后,引入地面运动目标相对背景杂波的稀疏先验概率分布,建立层级贝叶斯模型,应用变分贝叶斯期望最大算法实现合成孔径雷达地面运动目标成像的稀疏表示,同时对可能存在的目标高阶运动和载机运动误差造成的相位失调进行校正,以保证运动目标雷达图像的聚焦性能。仿真及实测数据的处理结果验证了应用该方法可以显著地提升多目标成像响应动态范围,相比传统方法具有明显的优越性。

When multiple ground moving targets are to be imaged simultaneously by a synthetic aperture radar, the dynamic range of the target responses in the SAR image will be reduced in terms of increased side-lobes. To this end, a parametric Bayesian learning algorithm is proposed in this paper for enhancing the sparse feature of the SAR image as well as reducing side-lobes of the target responses. First, the asymptotically linear Lv’s distribution as a novel time-frequency representation method is adopted to represent the Doppler parameters of the moving targets at the centroid frequency in the chirp rate domain. Accordingly, a quadratic Fourier dictionary is constructed for the following sparse Bayesian learning. Second, in order to evaluate the performance of the designated dictionary quantitatively, the mutual correlation among columns of the dictionary is calculated to evaluate the unaccessable restricted isometry property indirectly. Finally, by encoding a sparse prior or Laplacian distribution onto the multiple moving targets to be imaged, the Bayesian model is established in a hierarchical manner. Following variational Bayesian expectation maximization, the posterior of the target image can be analytically derived, and the sparse feature enhanced synthetic aperture radar image with a promising dynamic range in target response can be obtained. In addition, the non-systematic phase errors from both the airborne radar motion deviation and non-ideal target movement are considered within the Bayesian learning framework, which can therefore achieve promising results. The effectiveness of the proposed algorithm is validated by both simulations and raw data experiments, and the superiority is evaluated by comparing with conventional algorithms.