Automated Image Recognition of Staminate and Pistillate Flowers in Cucurbit Crops for Precision Pollination and Growth Management

Abstract

Cucurbit crops, such as watermelon, pumpkin, and cucumber, exhibit distinct monoecious characteristics, with significant differences in the timing and morphology of staminate and pistillate flowers. Staminate flowers are typically smaller, borne on shorter pedicels, and characterized by prominent anther structures, whereas pistillate flowers are larger, borne on longer pedicels, and distinguished by well-developed ovary structures. Image recognition technology enables automated flower identification and classification, assisting farmers in optimizing pollination strategies and crop management. In controlled environments, manual pollination remains a critical step for enhancing fruit quality and yield. Automated monitoring of flowering stages and floral abundance allows for precise control of production timing and output. Traditional manual pollination relies heavily on human experience, which is time-consuming, error-prone, and inefficient due to challenges in visual identification, time constraints, and precision requirements. By integrating an image-based recognition system, the identification of staminate and pistillate flowers can be automated, streamlining pollination processes, improving efficiency and success rates, and reducing labor and costs. Furthermore, flower image analysis contributes to pest and disease management by detecting early signs of plant health issues through morphological abnormalities. In this study, we established an image dataset based on the black-seeded mini watermelon cultivar, comprising 939 staminate flower images and 311 pistillate flower images. A YOLO v2 deep learning model was trained on this dataset, achieving an accuracy rate exceeding 97%. Future research will expand the database to include stem classification (main vine, secondary vine, and tertiary vine) to support the development of automated field operations for cucurbit crops.