Floral Precision: Investigating Pea (Pisum sativum L.) Flowering with High Throughput Field Phenotyping and Object Detection

Background Flowering is one of the most important and sensitive process throughout a plants life as it marks the start of the reproductive phase. Therefore, phenotyping the continuous development of flowering is crucial for crop breeding. For phenotyping, visual ratings have been a standard method for decades, to observe flowering dynamics by determining timepoints, such as start, end or duration. However, high throughput field phenotyping (HTFP) methods have emerged, providing an objective and efficient approach. We developed an approach that allows to collect detailed data not only about pea flowering dynamics, but additionally flower intensity (flowers per area). For this purpose, an object detection model, based on YOLOv8 was trained on RGB-images. The images were automatically acquired by the field phenotyping platform (FIP) of ETH Z¨urich for 12 pea breeding lines over two years. Results The trained model reached high accuracy for open flower detection, which allowed to monitor flower dynamics and intensity over time. Flower intensity throughout the development of the plants was highly correlated (R2= 0.967) to ground truth data taken in the field. Clear differences in timing, intensity of flowering and fruiting efficiency were detected between breeding lines and years. Furthermore, high correlation between maximal flower numbers and yield components such as seed amount were observed. Conclusion This automated, data-driven method of flower detection proved itself as a reliable tool. This is promising for the use of RGB imaging methods to objectively assess not only timing but also flower intensity. Flower intensity allows to predict seed amount and has therefore potential as selection trait in breeding program. In addition, fruiting efficiency could be included in breeding programs.