UAV-Based Detection of Deciduous Tree Species Using Structural and Spectral Characteristics

Abstract

The use of remote sensing technology is essential for identifying and mapping tree species. In species management, remote sensing tools like Unmanned Aerial Vehicles (UAVs) are used because of their short-cycle replication, high-resolution images, and 3D capabilities. The main objectives of this research were to evaluate the ability to use UAV images and the reliability of three Nearest Neighbor (NN), Random Forest (RF), and Decision Tree (DT) algorithms, as well as the ability to differentiate deciduous tree species based on their spectral and structural characteristics. UAV images were obtained and processed, and 3D canopy crown structure features, i.e. DSM, DTM, CHM, and mean slope of the canopy crown, were prepared for object-based classification. The results showed that adding the structural feature of CHM, DSM, and slope, as combined with multispectral bands, could improve the results compared to using only multispectral bands for NN and RF algorithms. However, the DT algorithm provided the highest classification accuracy with an overall accuracy of 69.04% and a Kappa coefficient of 0.595, using spectral characteristics of the main bands, vegetation indices, and texture analysis. In contrast to the DT algorithm, which does not improve classification results by using tree structural properties, CHM shape properties in combination with their spectral properties can improve classification results. Overall, in dense deciduous forests where all trees have normal spectral reflections during the growing season, UAV images and structural features such as mean slope provide valuable information.