[3Yin2] Interactive session 1

[3Yin2-35] Estimation of pseudo-color images from of monochromatic asteroid images using Generative Adversarial Network

〇Ren Hashimoto¹, Rintaro Kishi¹, Kimikazu Tanase¹, Rie Honda², ONC Team Hayabusa2 (1.Kochi University, 2.Ehime University)

Keywords:generative adversarial network, pseudo-color image, asteroid image, exploration

The asteroid explorer Hayabusa2 conducted imaging of the asteroid Ryugu with the Optical Navigation Camera (ONC). The data were usually taken as the visible 7 band images, which provided insight into the material distribution on the asteroid. However, most of the images taken at low altitudes have only one or three bands due to operational constraints, which makes material estimation difficult. A similar problem can occur in earth observation satellite imagery, where if no image is acquired by one sensor, it can be estimated from images taken by other sensors at the same time. In this study, with these applications in mind, we will conduct an estimation experiment of a high-resolution pseudo-color image from a high-resolution monochromatic image of the asteroid Ryugu and verify its usefulness. We applied pix2pix algorithm, which is one of Generative Adversarial Network (GAN), to artificially degraded images of ONC’s pseudo-color images and monochromatic images. The result shows that PSNRs for each color are the values ranging from 33 to 40 dB. We will report the validation result for the high-resolution images, and also the potential use of this method to detect unusual region on the planetary bodies.