MIMO Radar 3D Imaging Based on Combined Amplitude and Total Variation Cost Function With Sequential Order One Negative Exponential Form
MIMO Radar 3D Imaging Based on Combined Amplitude and Total Variation Cost Function With Sequential Order One Negative Exponential Form
复制标题
DOI：
10.1109/tip.2014.2311735
复制
发表时间：
2014-03
影响因子：
10.6
通讯作者：
Changzheng Ma;T. Yeo;Yongbo Zhao;Junjie Feng
中科院分区：
文献类型：
--
作者：
Changzheng Ma;T. Yeo;Yongbo Zhao;Junjie Feng
在逆合成孔径雷达（ISAR）成像中，目标通常被视为由少量强（镜面）散射体组成，并且这些强散射体在成像体积中的分布是稀疏的。在本文中，我们建议将稀疏信号恢复方法结合到3D多输入多输出雷达成像算法中。提出了在 ℓ1 和 ℓ0 范数之间形成同伦的序列一阶负指数 (SOONE) 函数来衡量稀疏性。梯度投影用于解决约束非凸 SOONE 函数最小化问题并恢复稀疏信号。然而，虽然梯度投影方法计算简单，但当算法中的矩阵病态时，它并不鲁棒。因此，我们进一步建议使用对角加载和奇异值分解方法来提高算法的鲁棒性。为了处理具有大平坦表面的目标，还提出了组合幅度和总变分目标函数来正则化平坦表面的形状。仿真结果表明，在高信噪比情况下，所提出的 SOONE 函数梯度投影方法在恢复 ±1 随机尖峰稀疏信号方面优于正交匹配追踪、CoSaMp、ℓ1-magic、拉普拉斯先验贝叶斯方法、平滑ℓ0 方法和 ℓ1-ℓs。使用新方法获得的模拟3D图像和真实数据ISAR图像的质量优于传统相关方法和最小ℓ2范数方法，并且与上述稀疏信号恢复算法具有竞争力。
In inverse synthetic aperture radar (ISAR) imaging, a target is usually regarded as consist of a few strong (specular) scatterers and the distribution of these strong scatterers is sparse in the imaging volume. In this paper, we propose to incorporate the sparse signal recovery method in 3D multiple-input multiple-output radar imaging algorithm. Sequential order one negative exponential (SOONE) function, which forms homotopy between ℓ1 and ℓ0 norms, is proposed to measure the sparsity. Gradient projection is used to solve a constrained nonconvex SOONE function minimization problem and recover the sparse signal. However, while the gradient projection method is computationally simple, it is not robust when a matrix in the algorithm is ill conditioned. We thus further propose using diagonal loading and singular value decomposition methods to improve the robustness of the algorithm. In order to handle targets with large flat surfaces, a combined amplitude and total-variation objective function is also proposed to regularize the shapes of the flat surfaces. Simulation results show that the proposed gradient projection of SOONE function method is better than orthogonal matching pursuit, CoSaMp, ℓ1-magic, Bayesian method with Laplace prior, smoothed ℓ0 method, and ℓ1-ℓs in high SNR cases for recovery of ±1 random spikes sparse signal. The quality of the simulated 3D images and real data ISAR images obtained using the new method is better than that of the conventional correlation method and minimum ℓ2 norm method, and competitive to the aforementioned sparse signal recovery algorithms.