AI-based Compression of SDO/AIA for Solar Coronal Features

In this study, we apply AI-based image compression methods to the Atmospheric Imaging Assembly (AIA) data on board the Solar Dynamics Observatory (SDO). We apply pretrained variational autoencoder (VAE) models as well as versions fine-tuned to solar extreme ultraviolet (EUV) data. We evaluate quality metrics such as peak signal-to-noise ratio (PSNR) and the multiscale structural similarity index measure (MS-SSIM) for human perception ability as a function of compression rates, represented by bits per pixel (bpp). Additionally, we analyze the differential emission measure (DEM) and pixel intensity variation of coronal features such as coronal loops and bright points and compare coronal loop oscillations between the original and decompressed data. We compare these results with conventional methods such as the JPEG method and the RICE algorithm commonly applied to solar spacecraft data. The major results from this study are as follows. First, AI-based image compression methods show higher MS-SSIM values than the conventional methods, at similar compression rates. Second, the DEM curves from the AI-based decompressed data generally follow the main trend of the original data. Third, the pixel intensity variations show wavelength-dependent differences, but the decompressed data generally retain the main trend of the original intensity variations. Fourth, the coronal loop oscillation obtained from the decompressed data is consistent with that of the observed coronal loop in the original data. We hope that AI-based image compression methods can effectively handle large amounts of solar data for future deep space missions and data storage.