Efficient ISAR Image Prediction Method for Multi-Target Through Deep Learning

In highly dynamic multi-target scenes, inverse synthetic aperture radar (ISAR) imaging has important effects on analyzing electromagnetic scattering characteristics of clustered targets, which are composed of the electromagnetic scattering characteristics of each single target and the coupling characteristics between targets. Considering that simulation calculation of the coupling characteristics is time-consuming, it cannot meet the needs of real-time calculation of multi-target ISAR images. To address the issue, this paper proposes a deep neural network (DNN) based on U-Net, which predicts the ISAR images of the multi-target scene from the ISAR image of each single target. Compared with physical optics (PO) simulation and range-Doppler (R-D) imaging methods, the proposed DNN can achieve a 99% speed increase while maintaining a mean square error (MSE) of 0.21.