DOI: 10.11118/978-80-7701-024-5-0047

Random Forest Algorithm and Convolutional Neural Networks for the Tree Species Classification in Remote Sensing Data

Zdeněk Patočka¹, Petr Strejček¹, Anton Malyshev¹

¹ Department of Forest Management and Applied Geoinformatics, Faculty of Forestry and Wood Technology, Mendel University in Brno, Zemědělská 1165/1, 613 00 Brno-Černá Pole, Czech Republic

This study deals with the classification of tree species using modern methods of machine- and deep-learning applied to satellite and drone date The aim of the study is to demonstrate the ability of these methods to accurately identify and classify different tree species.  The first part is focused on the use of DeepForest and Detectree2 algorithms for tree crown delineation, which allow efficient segmentation and detection of trees in complex aerial images. The work with YOLO (You Only Look Once) algorithm is presented, the purpose of which is to train a model for specific detection and classification of selected tree species from drone data. The results of this algorithm is compared to the results of Random Forest machine learning algorithm. Second part of the study is focused on tree species classification in the large area of the University Forest Area by using of the Sentinel-2 and PlanetScope data. It was used the Random Forest algorithm and permanent sample plot to train the algorithm a to create map of the main tree species.

Klíčová slova: deep learning, machine learning, multispectral data, drone, satellite, PlanetScope, Sentinel-2, Detectree2, YOLO

stránky: 47-52, online: 2025