Linear-Geometry Distortion Correction for Bistatic Inverse Synthetic Aperture Radar Imaging Based on Deep Learning Model

Compared with a monostatic inverse synthetic aperture radar (ISAR) imaging system, a bistatic ISAR (Bi-ISAR) system offers more comprehensive target information, along with higher system security and resistance to interference. However, the inherent geometric configuration of Bi-ISAR will introduce the linear-geometry distortion (LGD), causing the target to appear sheared in shape and impacting subsequent target recognition. To deal with this issue, in this paper, a novel deep learning-based algorithm is proposed to realize the LGD correction. In the proposed algorithm, a neural network called DoubleUNet is employed for the accurate semantic segmentation of target's ISAR image. Then the initial target ISAR image is transformed into two-dimensional point clouds based on the estimated radar parameters. Finally, the linear coupling relationship between azimuth and range dimensions is removed, beneficial to the target ISAR shape restoration and subsequent recognition. Simulation experiments validate the proposed algorithm.