1:45 PM - 2:00 PM
[U04-01] Tree Species Classification Using a UAV-Mounted Spectrometer
★Invited Papers
Keywords:Satellite, Remote sensing, Tree-species classification, Spectrometer, UAV
Optical remote sensing satellites provide a valuable opportunity to observe forest types, health, growth stages, and carbon sequestration on a global scale through spectral information. These observations play a crucial role in studying climate change and its impact on forest ecosystems.
However, commonly used satellite sensors capture spectral images with only a few broad bands, such as those used for NDVI and other vegetation indices. These indices are often insufficient for distinguishing inter- and intra-species variations in tree species. While hyperspectral satellite sensors offer a more detailed spectral resolution, they often suffer from low spatial resolution, limiting their applicability for precise tree species classification. Therefore, there remains a critical need to identify appropriate spectral bands for future remote sensing satellites that can effectively distinguish tree species and capture intra-species variations in diverse forest ecosystems worldwide.
To address this gap, this study aims to classify different tree species in various forest types using a UAV-mounted spectrometer measurements. Additionally, we attempt to explore the potential for assessing tree growth stages using spectral information in future research.
A simple spectrometer capable of capturing valuable spectral information with narrow bandwidth in the visible to near-infrared (VIS-NIR) range was used in this study. We conducted the UAV measurement in mixed dense forest in Ebetsu, Hokkaido, Japan, in 2024. For tree species classification, we used a Support Vector Machine (SVM) algorithm with three different feature sets: (1) all bands (188 bands), (2) principal components, and (3) selected bands. Our results show that eight common tree species in Hokkaido were classified with total accuracies of 80%, 81%, and 60%, respectively, for the three feature sets. These findings demonstrate that the UAV-mounted spectrometer can effectively capture spectral information.
In future work, we aim to expand our observations to different forest types to identify the most suitable spectral bands for improved tree species classification and investigate how spectral data can be utilized to assess tree growth stages using remote sensing techniques.
However, commonly used satellite sensors capture spectral images with only a few broad bands, such as those used for NDVI and other vegetation indices. These indices are often insufficient for distinguishing inter- and intra-species variations in tree species. While hyperspectral satellite sensors offer a more detailed spectral resolution, they often suffer from low spatial resolution, limiting their applicability for precise tree species classification. Therefore, there remains a critical need to identify appropriate spectral bands for future remote sensing satellites that can effectively distinguish tree species and capture intra-species variations in diverse forest ecosystems worldwide.
To address this gap, this study aims to classify different tree species in various forest types using a UAV-mounted spectrometer measurements. Additionally, we attempt to explore the potential for assessing tree growth stages using spectral information in future research.
A simple spectrometer capable of capturing valuable spectral information with narrow bandwidth in the visible to near-infrared (VIS-NIR) range was used in this study. We conducted the UAV measurement in mixed dense forest in Ebetsu, Hokkaido, Japan, in 2024. For tree species classification, we used a Support Vector Machine (SVM) algorithm with three different feature sets: (1) all bands (188 bands), (2) principal components, and (3) selected bands. Our results show that eight common tree species in Hokkaido were classified with total accuracies of 80%, 81%, and 60%, respectively, for the three feature sets. These findings demonstrate that the UAV-mounted spectrometer can effectively capture spectral information.
In future work, we aim to expand our observations to different forest types to identify the most suitable spectral bands for improved tree species classification and investigate how spectral data can be utilized to assess tree growth stages using remote sensing techniques.