面向天文多普勒差分测速的太阳/行星光谱对生成方法

In order to provide synchronous solar/planetary spectral pairs required for the celestial Doppler differential velocimetry，a VAE-Dual GAN，which combines variational auto encoder （VAE） with dual generative adversarial networks（Dual GAN）is proposed. Firstly，the measured solar spectrum is encoded into the latent space by VAE，which realizes the expansion from spectrum to spectral domain. Then，the latent space is mapped to the pseudo-planetary spectrum by Dual GAN. Finally，the reconstructed solar spectrum is generated by using the pseudo-planetary spectrum. Furthermore，the encoder and generator reconstruction losses are used to strengthen the constraints on the network. VAE-Dual GAN utilizes the conversion learning ability of Dual GAN to complete the transformation of two spectral domains and generate synchronous solar/planetary spectral pairs. The experimental results show that the VAE-Dual GAN generates high-quality solar/planetary spectral pairs，the accuracy of celestial Doppler differential velocity measurement is improved by more than 60%.