[3Xin4] Poster session 1

[3Xin4-24] A compression method of solar polarization data with autoencoder

〇Jargalmaa Batmunkh¹, Yusuke Iida¹, Takayoshi Oba², Haruhisa Iijima³ (1.Niigata University, 2.National Astronomical Observatory of Japan, 3.Nagoya University)

Keywords:astroinformatics, high-dimensional data, autoencoder

Solar spectral analysis plays an important role in solar physics research to understand the Sun-Earth relationship. Hinode Solar Optical Telescope (Hinode SOT/SP) has been accumulating solar spectro-polarimetry (SP) data for more than 15 years. However, processing this huge amount of high dimensional data is challenging even with the existing computational methods. To this end, we suggest a compressed representation of SP data using a deep learning technique that will be useful for further steps of solar spectral analysis, such as flare prediction, automatic categorization of spectra and detection of anomalous spectra. We built an autoencoder for compressing solar spectra containing Stokes I and V polarization parameters. The encoder converts the input (SP data) into a lower dimensional compressed representation of the spectra, and then decodes it back into the output (reconstruction). We compared performances of the model trained with different errors: standard loss as mean absolute error (mae), and customized loss as sum of weighted mae of Stokes I and V. From the scatter plot of true and reconstruction the model with customized loss function resulted in smaller standard deviations of 0.57-0.7% (continuums) and 2.71-3.16% (line centers) for Stokes I, and 4.79% (left line core) for Stokes V.