Tomato Flower Recognition and Localization Based on YOLOv8

In order to achieve efficient automated operation of tomato pollination robot, a flower identification and positioning method based on YOLOv8 keypoint detection and instance segmentation is proposed to address the problems of low accuracy of suitable pollination flower identification, fuzzy orientation information, and high cost of manual labelling. The YOLOv8 keypoint detection is used to screen out flowers with suitable pollination period and attitude and obtain the pistil key point, and further segment the screened flowers by instance segmentation to extract the flower shape centre, and then the computational analysis is used to achieve the flower orientation judgement and angle recognition, which provides the key information for the operation of the pollination robot. In the actual pollination experiment under the simulated plant factory environment, the flower recognition accuracy reaches 96.3%, the positioning error is less than 0.91 cm, and the orientation angle error is less than 8.8°. This method has high recognition accuracy, small positioning error and orientation angle error, and it can be used as a reference for the recognition of flower orientation and the solution of similar problems.