Comparative Analysis of Unmanned Aerial Vehicle Land Cover Classification of Two Study Sites in Serbia

In recent decades, using unmanned aerial vehicles (UAVs) for mapping and classifying the Earth's surface has surged. Ecological researchers increasingly rely on UAV imagery due to its centimeter-level resolution and rich information content compared to other remote sensing methods, such as satellite imagery. This study presents a comparative analysis of the land cover classification of orthomosaics in two distinct Serbian study sites using high-resolution imagery captured by different UAVs: the DJI Inspire 1 and DJI Phantom 4 Multispectral. In the study site Topli Do (Eastern Serbia), the DJI Inspire 1 equipped with RGB and NDVI modified cameras was utilized, generating the orthomosaic of six land cover classes. Random forest classification algorithms were employed achieving an overall accuracy of 92.46%. In the study site Glavica (Northern Serbia), aerial images were acquired using the DJI Phantom 4 Multispectral, featuring a multispectral sensor capturing data across blue, green, red, red-edge, NIR, and combined RGB ranges. Applying the random forest, five land cover classes were delineated with an accuracy of 95%. This finding underscores the transformative potential of multispectral cameras in enhancing land cover classification. Using spectral information captured in multiple bands has enabled determining plants up to the species level in heterogeneous vegetation types. Such classification represents a breakthrough in ecological monitoring, allowing scientists to gain insight into ecosystem dynamics, species distribution, and environmental change at the local scale.