Non-Cooperative Target Radar RCS Data Generation Based on Transfer Learning

Radar scattering cross section (RCS) is an important scattering property of radar targets and is crucial for radar target characterization, detection and identification. In non-cooperative target measurement, there are observation conditions, unknown attitude, limited measurement data and other difficulties. This paper proposes a non-cooperative target RCS data generation method based on adversarial discriminative domain adaptive generative network (ADDA - GN). The method combines deep generative network and transfer learning, takes a large number of simulation and darkroom measurement samples as source domain data, takes a small amount of measured data with unknown attitude angle as target domain data, and realises the generation of non-cooperative target full-attitude RCS characteristic data through the transfer learning from source domain to target domain. The validation results show that this method significantly improves the performance of the model in the target domain with the help of source-domain data transfer, and the generated target-domain RCS data is highly consistent with the real data, which provides a strong support for the study of electromagnetic scattering characteristics of non-cooperative targets and national defence applications.