SRGAN-LSTM-Based Celestial Spectral Velocimetry Compensation Method With Solar Activity Images

In celestial spectral velocimetry methods, when the solar spectra are stable, highly accurate velocity information of spacecraft can be derived from the Doppler frequency shift. However, solar activity can cause a frequency shift in the solar spectra, greatly affecting the accuracy of velocimetry. Due to limitations of spacecraft payload and computing resources, we propose a celestial spectral velocimetry compensation method that only uses single-frequency band solar images and solar spectra. In addition to a spectrometer used by traditional celestial spectral velocimetry methods, we only add one optical imaging sensor. We conduct a gray correlation analysis between single-frequency band solar image sequences and velocimetry bias induced by solar activity, which demonstrates a strong correlation between the two. Based on the above results, the velocimetry bias estimation can be modeled as a sequence regression problem. To address the regression issues, we develop a novel hybrid framework that combines semi-supervised regression generative adversarial networks (SRGANs) and long short-term memory (LSTM). The role of SRGAN is to improve the generalization ability of the regression model and extract features from the single-frequency band solar image sequences. LSTM receives the extracted feature sequences to capture temporal dependencies within the spectral sequences, thereby providing accurate velocimetry bias estimation. The experimental results indicate that SRGAN-LSTM accurately estimates the velocimetry bias caused by solar activity. After compensation, the velocity error decreases from 0.3242 to 0.0327 m/s.