2023 Fiscal Year Final Research Report
Recovery and Analysis of Satellite Data via Linear Programming and Deep Learning
| Project/Area Number |
20K21792
|
|---|
| Research Category |
Grant-in-Aid for Challenging Research (Exploratory)
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Medium-sized Section 60:Information science, computer engineering, and related fields
|
|---|
| Research Institution | National Graduate Institute for Policy Studies |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
上野 玄太 大学共同利用機関法人情報・システム研究機構(機構本部施設等), データサイエンス共同利用基盤施設, 教授 (40370093)
田中 未来 統計数理研究所, 数理・推論研究系, 准教授 (40737053)
池上 敦子 成蹊大学, その他部局等, 客員研究員 (90146936)
|
|---|
| Project Period (FY) |
2020-07-30 – 2024-03-31
|
| Keywords | 人工衛星データ / 深層学習 / プラズマ粒子 / 信号復元 |
|---|
| Outline of Final Research Achievements |
We developed deep learning neural network for recovery of plasma particle observation data from satellite Geotail. Observed plasma particles are categorized into 7 and 16 directions in latitude and longitude, respectively, and 32 levels in their energies. 3d histogram data of plasma particles in 7*16*32=3584 bins based on aforementioned categorization is transmitted to a ground station every 12 seconds for 8 hours a day. For the rest 16 hours, 2d histogram data of 16*32 obtained by marginalizing latitude counts is sent due to transmission limit. Human observers analyzes the data by using a chart called E-t spectrogram computed from 3d histogram. We developed a system to recover 3d histogram from marginalized 2d histogram by deep learning neural networks. Utilizing U-net, Res-net, Vanilla GANs with some modification, we were able to obtain fairly good estimation of E-t spectrogram.
|
| Free Research Field |
統計数理・数理工学
|
| Academic Significance and Societal Importance of the Research Achievements |
人工衛星 Geotail のプラズマ粒子観測データについて，深層学習を用いて，２次元ヒストグラムデータから３次元ヒストグラムデータを復元することを試みた．２次元ヒストグラムは全体の2/3を占めており，本研究成果により，今後これらのデータから３次元ヒストグラムを復元して，興味深い現象とされる，磁気リコネクションが起こっている様子などを知ることができると考えられる．
|
