Capsicum Flower Identification for Robotic Pollination in Greenhouses

Due to restrictions on bees and the lack of wind in the greenhouse, pollination could be a labor-intensive activity. Hence, an automated pollination process is preferred to improve the productivity in greenhouse settings. This paper introduces an image-based pollination method applicable within the greenhouse. A stereo-vision camera and a You Only Look Once (YOLO)-based image processing technique are employed to identify and locate the pollination-ready capsicum flowers in the greenhouse. The detected flower's location is communicated to a robotic system for it to be maneuvered in front of the flower to finish the required pollination. When tested on the test dataset using Precision & Recall Curve (PRC), the proposed detection method achieves an average detection precision of 0.76 for the first class (CapFlower) and 0.61 for the other (Bud).