Image Mask Constraints Based Radar Data Intelligent Generation Model

Abstract

Aiming at the problems of limited number of electromagnetic real data and insufficient features in radar target intelligent identification, a radar data intelligent generation model based on image mask constraints is proposed to realize the effective expansion of radar Range Doppler (R-D) images. First, a set of radar 1-D echo data of non-cooperative ship targets under real sea conditions is acquired. The velocity information is well preserved by using Fourier transform in the signal preprocessing stage, and the interactive program is designed to perform image annotation to form an image dataset. Then, a GAN-based data generation network model is built to take the image dataset as input, and a multi-scale constraint based on generalized regularization technique is designed to improve the similarity between the generated image and the real image to solve the problem of insufficient samples. Finally, an image mask is introduced into the generator to specify the generation location of target, and at the same time to improve the sample diversity during the data generation process. The dataset generated by the proposed method takes into account the similarity and diversity, and can provide data guarantee for the subsequent high-precision recognition of radar signals under small sample conditions