Japan Geoscience Union Meeting 2018

Presentation information

[JJ] Oral

A (Atmospheric and Hydrospheric Sciences) » A-CG Complex & General

[A-CG40] Material Circulations in Land Ecosystems

Thu. May 24, 2018 10:45 AM - 12:15 PM 106 (1F International Conference Hall, Makuhari Messe)

convener:Tomomichi Kato(Research Faculty of Agriculture, Hokkaido University), Takashi Hirano(Research Faculty of Agriculture, Hokkaido University), Hisashi Sato(海洋研究開発機構 地球表層物質循環研究分野, 共同), Ryuichi Hirata(National Institute for Environmental Studies), Chairperson:Sato Hisashi(JAMSTEC)

12:00 PM - 12:15 PM

[ACG40-06] Re-invent remote sensing by using deep learning

*Takeshi Ise1,2 (1.FSERC, Kyoto University, 2.PRESTO, JST)

Keywords:deep leaning, neural network, remote sensing

Remote sensing has been contributed to the nondestructive observations of the earth surface. Remote sensing can cover large areas with homogeneous observation. However, essentially, the conventional remote sensing technique was zero-dimensional, where spatial information such as textures of the earth surface was neglected.

Machine learning using deep neural networks (DNN) is the powerful tool for detecting objects, such as human faces, cars, dogs, etc. However, DNN has not been widely utilized to identify amorphous objects such as vegetation types in remotely sensed data.

In this study, using DNN, the method to classify vegetation types has been developed. The newly introduced method “chopped picture method” showed a good performance for classification of earth surfaces according to vegetation types, such as orchards, bamboo stands, etc. This can be the re-invention of remote sensing, because we now have a method to tap onto the rich information, the two-dimensional spatial information, in remotely sensed data.