基于复数深度神经网络的逆合成孔径雷达成像方法

Inverse Synthetic Aperture Radar Imaging Method Using Complex Value Deep Neural Network

压缩感知(Compressive sensing,CS)理论框架下逆合成孔径雷达(Inverse syntheitic operture radar,ISAR)成像的结果具有超分辨、无旁瓣干扰等特点,但CS ISAR成像方法性能仍然受到稀疏表示不准确和图像重建方法效率低等限制。基于深度神经网络(Deep neural network,DNN)的欠采样或不完整信号重建方法取得了瞩目的表现。DNN能够自主学习最优网络参数并挖掘出输入数据的抽象高层特征表示,但目前已有的DNN都为实数域的模型,无法直接用于复数形式数据处理。为了利用DNN的优势提高ISAR欠采样数据成像的质量,本文通过级联不同类型的复数网络层的方式,构建具有多级分解能力的复数深度神经网络(Complex value DND,CVDNN),利用CV‑DNN实现ISAR成像。实验结果表明,基于CV‑DNN的ISAR成像方法在成像质量和计算效率方面都优于传统压缩感知成像方法。

The results of the inverse synthetic aperture radar(ISAR) imaging in the framework of compressive sensing(CS) have the advantages of super resolution and no sidelobe interference. But the availability or appropriateness of the sparse representation of the target scene and the relatively low computational efficiency of image reconstruction algorithms limit the performance and application of the CS based ISAR imaging methods. Recently, the deep neural network(DNN) based under-sampled or incomplete signal reconstruction method achieve remarkable performance. DNN can extract the abstract feature representation from input data with the hidden layers and nonlinear activation layer. However,the existing DNNs are real domain models,and cannot be directly used in complex data processing. A complex value DNN(CV-DNN)with multistage decomposition ability is constructed by cascading different types of complex value network layers. Then,the CV-DNN is used for ISAR imaging. The CV-DNN architecture can extract and exploit the sparse feature of the target image extremely well by multi-layer nonlinear processing. The experimental results show that the proposed CV-DNN based ISAR imaging method can provide better shape reconstruction of target than state-of-the-art CS reconstruction algorithms and improve the imaging efficiency obviously.