Urban green space management is essential for sustainable urban planning, requiring accurate and efficient tree detection methods. This study compares three deep learning models—YOLOv11, faster region-based convolutional neural network (R-CNN), and DeepForest—for palm tree detection in an urban environment using high-resolution unmanned aerial vehicle (UAV) imagery. A UAV-based photogrammetric workflow was implemented, generating georeferenced orthomosaic maps to serve as the primary dataset. The models were assessed based on accuracy, processing speed, and suitability for urban vegetation mapping. YOLOv11 outperformed the other models, achieving an overall detection accuracy of 94.2% with an intersection over union (IoU) of 88.6%, making it the most suitable model for large-scale urban tree monitoring. Faster R-CNN demonstrated high precision (90.9%) but suffered from computational inefficiencies, leading to slower inference times. DeepForest, initially developed for forestry applications, exhibited the lowest accuracy (88.0%) due to false positives in dense urban settings. The statistical analysis confirmed that YOLOv11's Fr_score (93.9%) and recall (93.7%) were significantly higher than the competing models, making it the optimal choice for real-time urban tree monitoring. The study highlights the advantages of UAV-based deep learning applications for urban tree detection and underscores the importance of model selection based on environmental complexity and computational constraints.