基于CV-FISTA网络的稀疏孔径ISAR成像方法

逆合成孔径雷达(inverse synthetic aperture radar,ISAR)在雷达目标识别、空间监视和弹道导弹防御等领域发挥着重要作用。针对传统稀疏孔径ISAR成像算法对参数敏感和收敛速度慢的问题,提出一种基于复值快速迭代收缩阈值算法网络的稀疏孔径ISAR成像恢复方法。将加速近端梯度方法引入稀疏重构算法中,并将其迭代步骤构建为深度展开网络的隐藏层,构建初始参数相同的随机散射点和飞机散射点的数据集,将复值一维距离像作为网络的输入,利用ISAR像对应的标签对网络进行训练和验证。该方法直接处理复数数据替代传统的分实虚部两路计算方法,显著减少了计算负担。仿真实验表明,相较于传统模型驱动算法,通过对网络进行训练避免了手动调参过程,收敛速度更快,成像质量更高,而且对于特征差异较大的数据具有更好的泛化能力。

Sparse aperture ISAR imaging method based on CV-FISTA network

Inverse synthetic aperture radar(ISAR) plays an important role in radar target recognition, space surveillance, and ballistic missile defense. Considering that the traditional sparse aperture ISAR imaging algorithms are extremely sensitive to parameters and have slow convergence speed, this paper proposed a sparse aperture ISAR imaging recovery method based on complex valued-fast iterative shrinkage thresholding algorithm (CV-FISTA) network. This method first introduced the accelerated proximal gradient method into the sparse reconstruction algorithm and constructd its iterative steps into a hidden layer of the deeply unfolded network. Then, a dataset of random scattering points and aircraft scattering points with the same initial parameters was constructed, the complex one-dimensional distance image was used as the input of the network that was trained and verified by using the corresponding label of the ISAR image. Defterent from the traditional two-way calculation method including real and imaginary parts, this method directly processes comptex data, and therefore significantly reduces the computational burden. Compared with the traditional model-driven algorithm, simulation experiments verify that the proposed method can avoid setting parameters manually by training network, and have faster convergence speed, higher imaging quality, and better generalization ability for data with large feature differences.