Generating Radar Track Using Cycle-consistent Generative Adversarial Network

Abstract

The real radar track datasets of marine targets in complex environment are insufficient, and it is difficult to obtain a large number of datasets. To solve these issues, we propose a marine targets radar track data generative method, which is based on a cycle-consistent generative adversarial network. This method achieves the end-to-end mapping from the truth trajectory to the true error track using unpaired datasets. To make the generator better recover the details of the track, we creatively put the U-Net architecture to the generator of the network, where the U-Net skip-connection structure is used to ensure the effective transmission of features. The experimental verification was performed with a self-built dataset. The results show that the proposed method can effectively learn the error distribution of samples, generate track data with higher similarity to the true error distribution, and improve the accuracy of track classification.