Development of Real-time Female Flower Localization for Automated Pollination of Oriental Melon

Automation in agriculture is increasingly viewed as a solution to current challenges, such as labor shortages, shifting consumer preferences, and ecological responsibility. Within agricultural automation, object detection and location estimation play crucial roles in developing automated task systems for the cultivation process. In this study, we aim to develop a female flower detection model and estimate the three-dimensional location for an automated pollination system for oriental melons. First, we employ real-time object detection models, YOLOv5 and YOLOv8, to identify female flowers and compare their performance in terms of detection accuracy and processing speed. Both YOLOv5 and YOLOv8 achieved high accuracy, with mAP50 scores exceeding 0.85. However, YOLOv8 demonstrated superior real-time processing capability. Second, we utilize depth information obtained from a stereo vision camera to estimate the three-dimensional locations of the flowers detected by the object detection model. To this end, we built a ROS (Robot operating system) based female flower localization system and analyzed the localization data for female flowers of oriental melons.