3:30 PM - 5:00 PM
[PPS06-P15] DEMs with super-resolution and photo-geological maps by machine learning to install on Web-GIS
Keywords:Moon, GIS, Machine Learning
USA has led Gateway which is a small space station in orbit on the Moon as a successor to the International Space Station. Space agencies and industries have been involved. The Artemis program with the Gateway aims to land on the Moon and to develop its surface. The program has remarked permanently shadowed polar regions, because we expect water resource. We need precise polar maps, because the regions show restrictions to be selected as candidates of solar-power plant and to be accessed by land.
The polar regions show many shadings and shadows (lack of observations) on image data, and footprint concentrations of laser altimeter. Although DEMs derived from laser altimeter have poor resolution compared with images, the DEM would be useful by their super-resolution. The first purpose is to accomplish it with machine learning. Second purpose is to substitute traditional mapping by hand with semantic segmentation and classification by machine learning, which is similar to the case of Google map. Finally, the third purpose is to design, to develop, and to implement GIS with API for displaying our data products. This research has established not only new data products but also Web-GIS. We will report the results in this poster session.
In the super-resolution part, this research has accomplished to generate high-resolution DEM (2m/pixel) from low-resolution DEM (8m/pixel) by SRGAN. The generated DEM shows topographical features more detailed than LOLA GDR in 5 m/pixel. We compared two approaches in super-resolution. One is SRGAN [1], another is pix2pix [2]. We have selected SRGAN because of its precision. The generated DEM has no lack of observations due to shadows. Details of the geological mapping will be reported as a poster in the session [P-PS06 Lunar Science and Exploration], “Automatic geological Mapping by Semantic Segmentation of KAGUYA Multiband Images Using Deeplabv3+” by Goto. Details of the GIS developing will be reported as a poster in the session [P-PS06 Lunar Science and Exploration], “GIS development for the Moon by deck.gl” by Ibuka.
References:
[1] Ledig et al. (2017) IEEE Conference on Computer Vision and Pattern Recognition (CVPR),681-4690 “Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network”
[2] LOLA GDR http://imbrium.mit.edu/BROWSE/LOLA_GDR/
[3] Ogino et al. (2023) 54th LPSC “Verification of Super-Resolution Method for Lunar Polar DEM by Generative Adversarial Networks”
[4] Ibuka et al. (2023) 54th LPSC “Image-To-DEM Translation with Conditional Adversarial Networks of Depth Estimation Based on Monocular Images on The Moon”
The polar regions show many shadings and shadows (lack of observations) on image data, and footprint concentrations of laser altimeter. Although DEMs derived from laser altimeter have poor resolution compared with images, the DEM would be useful by their super-resolution. The first purpose is to accomplish it with machine learning. Second purpose is to substitute traditional mapping by hand with semantic segmentation and classification by machine learning, which is similar to the case of Google map. Finally, the third purpose is to design, to develop, and to implement GIS with API for displaying our data products. This research has established not only new data products but also Web-GIS. We will report the results in this poster session.
In the super-resolution part, this research has accomplished to generate high-resolution DEM (2m/pixel) from low-resolution DEM (8m/pixel) by SRGAN. The generated DEM shows topographical features more detailed than LOLA GDR in 5 m/pixel. We compared two approaches in super-resolution. One is SRGAN [1], another is pix2pix [2]. We have selected SRGAN because of its precision. The generated DEM has no lack of observations due to shadows. Details of the geological mapping will be reported as a poster in the session [P-PS06 Lunar Science and Exploration], “Automatic geological Mapping by Semantic Segmentation of KAGUYA Multiband Images Using Deeplabv3+” by Goto. Details of the GIS developing will be reported as a poster in the session [P-PS06 Lunar Science and Exploration], “GIS development for the Moon by deck.gl” by Ibuka.
References:
[1] Ledig et al. (2017) IEEE Conference on Computer Vision and Pattern Recognition (CVPR),681-4690 “Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network”
[2] LOLA GDR http://imbrium.mit.edu/BROWSE/LOLA_GDR/
[3] Ogino et al. (2023) 54th LPSC “Verification of Super-Resolution Method for Lunar Polar DEM by Generative Adversarial Networks”
[4] Ibuka et al. (2023) 54th LPSC “Image-To-DEM Translation with Conditional Adversarial Networks of Depth Estimation Based on Monocular Images on The Moon”