实孔径超分辨技术已经广泛应用于雷达前视成像领域,然而其中大部分迭代求解方法往往面临参数选择困难和迭代重建耗时等问题。对此,文中提出了一种基于交替方向乘子法网络的机载雷达前视成像方法。该方法将前视成像构建为施加了稀疏约束的解卷积问题,把交替方向乘子法(ADMM)分离变量迭代求解的过程映射成一个深度神经网络,即ADMM-Net (ADMMN)。经过训练,ADMMN 可以在有限的网络深度下学习最优的参数,借此提高雷达方位向的分辨率。实验结果表明,相较于传统迭代算法,ADMMN 可以用更少的时间实现前视超分辨成像。

Real aperture super-resolution technology has been widely used in the field of radar forward-looking imaging, but most of the current iterative methods face the problems of difficult parameter selection and time-consuming iterative reconstruction. In this paper, a forward-looking imaging method of airborne radar based on alternating direction multiplier network is proposed. In this method, the forward-looking imaging is constructed as a deconvolution problem with sparse constraints, and the iterative solution process of separating variables by alternating direction multiplier method (ADMM) is mapped into a deep neural network, namely ADMM-Net (ADMMN). After training, ADMMN can learn the optimal parameters under limited network depth, so as to improve the azimuth resolution of radar. Experimental results show that, compared with the traditional iterative algorithm, ADMMN can achieve super-resolution forward-looking imaging in less time.

TN957