Small Sample Target Recognition Based on Radar HRRP and SDAE-WACGAN

Abstract

The high resolution range profile (HRRP) of radar targets is commonly used for target recognition, and the recognition of non-cooperative targets is one of the urgent problems to be solved in radar target recognition. In order to address the noise impact and small sample issues in non-cooperative target recognition, this paper proposes a radar target HRRP recognition method based on SDAE-WACGAN. This method combines Stacked Denoising Auto-encoders (SDAE) and Weighted Auxiliary Classifier Generative Adversarial Networks (WACGAN). In this networks, the decoder of SDAE is used as the generator of WACGAN and the weight coefficients is introduced on the basis of ACGAN, so that the network can generate high-quality data that is more consistent with the real sample distribution, and can be more robust to noise. Experimental results show that compared with other commonly used models, the proposed method achieves higher recognition accuracy in scenarios with small samples and high noise, and demonstrates certain advantages in different SNRs and different number of sample sets.

Keywords

• High resolution range profile (HRRP)
• non-cooperative target recognition
• stacked denoising auto-encoders (SDAE)
• weighted auxiliary classifier generative adversarial networks (WACGAN)