[M-TT46] Introducing metaverse to agriculture. Are we ready?

[MTT46-P05] Preliminary Study of Machine Learning-based Tree Species Discrimination using Spectrometer Data Measured at Uryu Research Forest, Hokkaido

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

Keywords:Tree Species Classification, Machine Learning, Remote Sensing

Several types of tree species such as a Timber tree are considered as forest resources. Therefore, specific trees species are needed to be classified accordingly for the process of cutting down can be done properly and used according to the market needs. Manual identification to determine each species in an area or using remote sensing spectral data are examples of a method to collect accurate information about the forest. Nevertheless, manual identification requires high labor costs and remote sensing data relies on specific indices to discriminate the species. The field experiment was done at a specified area in the Uryu Research Forest, Uryu-gun, Hokkaido in collaboration with a team from the Graduate School of Engineering, Hokkaido University. The targeted area consists of 14 species of trees. The tree locations are scattered within the area and this creates much more difficulty to discriminate the species. Here we show that each species can be distinguished using a wavelength range of 520 - 560 nm and 720 - 760 nm. However, the spectral distribution is still susceptible to sunlight intensity. This error needs to be solved before the spectral results are to be fed into the training sets for the support vector machine (SVM) algorithm. It is also found that the flight path of the drone and the location of the known tree species are not well-positioned. These differences led to the labeling process of the measured spectral and the tree is based on assumptions. As a result, the number of training data sets for each species is very low and this later can affect the classification accuracy. It is expected that the classification can be done successfully using a spectrometer with a lack of spatial resolution but providing good classification accuracy compared to its superior instrument such as spectrophotometer (spectral imaging).