5:15 PM - 6:45 PM
[U04-P03] Classifying tree species using a simple spectrometer and 4-band camera
Keywords:Tree-species, Classification, Hyperspectral image , 4 band camera
For long-term forest monitoring and climate change studies, accurate estimation of forest carbon sequestration from biomass is important. In the last few decades, active and passive remote sensing techniques have played an important role in this field of study. Hyperspectral imagery, synthetic aperture radar (SAR) and lidar 3D imagery can provide sufficient data for accurate classification of forest tree-species. However, the acquisition of large amounts of data is costly and time consuming. Finding the best way to obtain valuable data with low-cost technologies will help to improve tree species classification and, consequently, forest biomass estimates.
We performed the classification of mixed temperate forest using time-series hyperspectral imagery acquired with a simple motorized spectrometer and 4-band camera. The HSI measurement was carried out from May 2023 to October 2023, including the peak and autumn seasons. The spectrometer was placed inside the building, measuring the spectrum through the window, and we assumed that using all 210 bands of the HSI for classification is unnecessary. Therefore, one of our goals is to find the bands that best correlate with species classification. To check the correlation, 4 bands are selected and measured depending on the previous year's measurement.
According to our results, the selected 4 bands worked well in the fall season, but were not good in the peak season of the forest. Moreover, several machine learning algorithms were used to distinguish tree-species. As a result, we presented the strategy of obtaining valuable data using multiple measurements. The important factors that can affect the measurement and its results were evaluated and discussed. Our future goal is to improve the estimation of forest biomass using tree species information and individual tree growth.
We performed the classification of mixed temperate forest using time-series hyperspectral imagery acquired with a simple motorized spectrometer and 4-band camera. The HSI measurement was carried out from May 2023 to October 2023, including the peak and autumn seasons. The spectrometer was placed inside the building, measuring the spectrum through the window, and we assumed that using all 210 bands of the HSI for classification is unnecessary. Therefore, one of our goals is to find the bands that best correlate with species classification. To check the correlation, 4 bands are selected and measured depending on the previous year's measurement.
According to our results, the selected 4 bands worked well in the fall season, but were not good in the peak season of the forest. Moreover, several machine learning algorithms were used to distinguish tree-species. As a result, we presented the strategy of obtaining valuable data using multiple measurements. The important factors that can affect the measurement and its results were evaluated and discussed. Our future goal is to improve the estimation of forest biomass using tree species information and individual tree growth.