Semantic Segmentation in Unmanned Aerial Vehicle Surveillance for Detailed Urban Tree Mapping and Species Classification

Urban greening is becoming increasingly important, which has increased interest in tracking and analyzing urban vegetation and highlighted the necessity for an accurate and practical evaluation of greenery inside cities. Unmanned Aerial Vehicles (UAVs) have become an essential tool for mapping vegetation at a broad scale, frequently using orthographic views to manage the heavy workload. This research examines semantic segmentation and a particular issue with tree species identification brought on by perspective views from different heights. The methodology involved the semantic segmentation methods used without orthorectification on a single low-altitude (<100 m) UAV RGB image. It suggested that an innovative mathematical model be used to successfully reduce GPS inaccuracies, which are known to impact the precision of tree species identification. The proposed system demonstrated an overall accuracy of 89.81 (±1.45) %. Furthermore, the mathematical model exhibits the potential to significantly reduce GPS inaccuracies, leading to enhanced precision in tree species identification.