UAV-based YOLOv11 system for automated detection and counting of male hemp plants bearing staminate flowers

Abstract The flowering stage is a crucial phenological phase in hemp production, determining the maturity group of grain and fiber varieties, yield potential, and optimal harvest timing. Enabling early-stage detection of flowering male plants and distinguishing them from non-flowering or female plants offers a quantitative solution for reproductive monitoring, maturity assessment, sex-ratio estimation, and crop management. Automated flower detection offers a practical alternative to labor-intensive and error-prone manual scouting; however, accurate identification remains challenging due to the small size of floral structures, dense canopies, occlusion, and variable lighting conditions. Therefore, this study aimed to develop a practical model to accurately detect and count male plants at the flowering stage under diverse field conditions. We developed a YOLO11-based drone imaging system for real-time detection of flowering plants, which outperformed earlier versions in identifying small and occluded floral structures while maintaining competitive inference speed. Model predictions showed strong agreement with manual counts of plants bearing male flowers, explaining 95% of the variation (R 2 = 0.95). The YOLOv11 model achieved 76.5% precision, 78% recall, [email protected] of 80%, and an F1 score of 77% under field conditions. The application is publicly accessible via web and mobile interfaces, allowing users to upload field images and receive real-time counts of male plants in flower. Accurate flower detection and timing estimation will support farmers' harvest-time decisions and reduce the workload of hemp researchers during phenotyping.