Visibility Extension of 1-D Aperture Synthesis by a Residual CNN for Spatial Resolution Enhancement

Abstract

In order to improve the spatial resolution of a one-dimensional aperture synthesis (1-D AS) radiometer without increasing the size of the antenna array, the method of visibility extension (VE) is proposed in this article. In the VE method, prior information about the visibility distribution of various scenes is learnt by a residual convolutional neural network (ResCNN). Specifically, the relationship between the distribution of low-frequency visibility and that of high-frequency visibility is learnt. Then, the ResCNN is used to estimate the high-frequency visibility samples from the low-frequency visibility samples obtained by the AS system. Furthermore, the low- and high-frequency visibility samples are combined to reconstruct the brightness temperature image of the scene, to enhance the spatial resolution of AS. The simulation and experiment both demonstrate that the VE method can enhance the spatial resolution of 1-D AS.