基于改进CycleGAN的ISAR图像超分辨方法

图像超分辨是解决ISAR欺骗干扰中由于模型样本不完备导致难以对大带宽ISAR实现高逼真假目标模拟的重要手段。利用生成对抗网络（GAN）可通过端到端映射实现ISAR图像的超分辨，然而，当测试输入样本与训练输入样本分辨率差异较大时，超分辨图像中会出现伪散射点从而导致目标失真。考虑到循环生成对抗网络（CycleGAN）对输入样本差异适应性较好，本文提出了一种基于改进CycleGAN的ISAR欺骗干扰超分辨样本生成方法，分别从损失函数、优化过程、判别器结构三方面对CycleGAN网络结构进行改进，加快了网络的收敛速度，同时对于输入分辨率差异较大的ISAR图像泛化性能更好。利用暗室测量数据验证了所提方法的有效性，与GAN方法相比，对于训练输入样本分辨率差异较大的测试输入样本，生成的超分辨样本散射点位置与真实数据具有更好的匹配效果。

ISAR Images Super-Resolution Method Based on Ameliorated CycleGAN

? ?For deception jamming against ISAR， it is difficult to generate high-fidelity fake targets due to the lack of model samples. To address this problem， image super-resolution achieved by end-to-end mapping learned by generative adversarial network can be employed. However， when the resolutions of the training set and the test set are quite different， fake scattering points will appear in the super-resolution results，which may led to the target distortion. In this paper， we propose a super-resolution sample generation method based on ameliorated CycleGAN due to the robustness of CycleGAN to the diversity of input samples. Compared with the original CycleGAN， we modify loss function， optimization process and framework of discriminator， which speeds up the convergence and enhances the network’s generalization performance for different resolution input ISAR images. Relevant experiments using data tested in anechoic chamber demonstrate the effectiveness and superiority of our method compared to GAN. Especially when the resolution of the training and test sets are very different， super-resolution image generated by our method has more matching scattering points with ground truth.