Japan Geoscience Union Meeting 2022

Presentation information

[E] Poster

M (Multidisciplinary and Interdisciplinary) » M-SD Space Development & Earth Observation from Space

[M-SD41] Geospatial applications for natural resources, environment and agriculture

Sun. May 29, 2022 11:00 AM - 1:00 PM Online Poster Zoom Room (29) (Ch.29)

convener:Abdul Rashid Bin Mohamed Shariff(Universiti Putra Malaysia ), convener:Yukihiro Takahashi(Department of Cosmosciences, Graduate School of Science, Hokkaido University), Chairperson:Abdul Rashid Bin Mohamed Shariff(Universiti Putra Malaysia), Anuar Ahmad(Universiti Teknokogi Malaysia), Nobuyasu Naruse(Faculty of Medicine, Shiga University of Medical Science)

11:00 AM - 1:00 PM

[MSD41-P02] Spectral Measurement using Drone at Uryu Experimental Forest for Tree Species Classification

*Ahmad Shaqeer Mohamed Thaheer1, Yukihiro Takahashi1 (1.Hokkaido University)

Accurate information about forest tree species composition is essential for sustainable forest management. In this study, a spectrometer is attached to a drone and flew over a targeted area in a research forest in Uryu-gun, Hokkaido, Japan. The field measurement was done in collaboration with a team from the Graduate School of Engineering, Hokkaido University. The targeted area consists of 14 species of trees. The drone measurement is done twice with different yaw angle to provide different measurement angle. The spectrometer provides a high spectral resolution data, a RGB image and an image timestamp for the analysis. Meanwhile the drone recorded the coordinate of flight path and the flight timestamp. Later, both timestamps as well as the drone coordinates are used to find the image coordinates. The coordinates is then correlated with a known coordinates of the tree species. This process is to match the spectral data and the tree and correctly labelled it for training datasets. It is found that, each species consists of one or two spectral data that is perfectly aligned with the coordinates of the known tree species. This data is found to be very low in order to define it as training datasets. The other spectral data positions are found to be marginally away from the actual tree locations and can be assumed to be the tree spectral data to solve the low number of training datasets. Finally, the spectral data is processed to obtain the reflectance factor and later fed into support vector machine (SVM) algorithm for tree species classification.