Development of intelligent earth observation system based on incomplete data reconstruction and automatic selection of training samples

2020 Fiscal Year Final Research Report

• Project/Area Number: 18K18067
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: The University of Tokyo (2020); Institute of Physical and Chemical Research (2018-2019)
• Principal Investigator: Yokoya Naoto 東京大学, 大学院新領域創成科学研究科, 講師 (40710728)
• Project Period (FY): 2018-04-01 – 2021-03-31
• Keywords: データ融合 / 画像合成 / 3次元変化検出 / 災害状況把握

Outline of Final Research Achievements

We developed methods for image reconstruction and image recognition that overcomes data incompleteness by complementarily utilizing the strengths of heterogeneous sensors in Earth observation. We also studied how to efficiently collect and generate training data from auxiliary data such as ground-shot images and simulation data. We applied our methods to disaster damage assessment, where data incompleteness and lack of training data are serious problems, and realized all-weather change detection using heterogeneous images from two different time periods and rapid estimation of 3D change of the ground surface, which were difficult with conventional techniques.

Free Research Field

知覚情報処理

Academic Significance and Societal Importance of the Research Achievements

本研究では，異なるセンサから得られる地球観測データを用いた変化検出を可能としたことと，教示データの収集が困難な３次元変化の高速推定を実現したことに大きな意義がある．これらの成果により，観測データの異種性や教示データの不足がボトルネットとなっていた災害状況把握において，浸水深や土石流による地形変化，建物の被害状況など，災害対応分野で必要とされる災害情報を迅速に提供することが可能となったことが最大の成果である．