Annotated dataset for YOLO models for Date palm tree detection

Date palm cultivation is a cornerstone of the UAE’s cultural heritage and economy, supporting nearly 40 million trees across diverse arid landscapes. However, pests, diseases, and environmental stress threaten yield stability. Accurate, large-scale detection and monitoring of individual palms remain challenging due to complex plantation structures and harsh conditions. This study leverages advances in deep learning and open-source satellite imagery to develop a scalable, cost-effective approach for precise date palm identification and enumeration. We designed an AI-driven framework employing the YOLOv11-Medium model trained on 215 manually annotated and augmented satellite images from Google Earth, covering diverse plantation types and environments. Utilizing Roboflow for annotation and dataset generation, the model incorporated advanced multi-scale feature extraction and spatial attention modules. Training spanned 100 epochs on 15084 labelled instances, with five different classes viz., Field crops, Fruit Trees, Greenhouses, Other vegetation and Date palm trees with comprehensive validation and comparative evaluation against recent YOLO variants. YOLOv11-Medium with transfer learning achieved outstanding performance with 90.3% mAP@50, and an accuracy of 79% outperforming YOLOv8, YOLOv12, and YOLO-NAS. Precision Recall, F1score and Accuracy indicate the superior performance of YOLOv11-Medium model in object detection of date palm trees. Qualitative assessment showed high accuracy in both sparse and dense plantations, with minor limitations in overlapping canopies, multiple classes in a parcel and shaded areas. Our YOLOv11-Medium trained with transfer learning demonstrates a robust, efficient solution for automated large-scale date palm monitoring in arid regions. Its architectural innovations and stable training performance highlight its potential for integration into precision agriculture.