Machine learning assisted analysis of rice flower opening times using a low-cost time-lapse camera.

Abstract

Flower opening time (FOT) is a key trait for successful reproduction and reproductive isolation. In crop science, FOT is critical for stress avoidance and efficient breeding practices. This study developed a system for the automatic detection of rice flower openings and FOT estimation by integrating a low-cost time-lapse camera with machine learning technology. This approach enabled high-resolution monitoring of flowering dynamics in two cultivars: the japonica cultivar Taichung 65 (T65) and the indica cultivar IR24. The system accurately identified regions containing open flowers, and the estimated FOTs varied within a 3-h range, with a root mean square error of approximately 30 min compared to manual detection. A significant difference in estimated FOTs between IR24 and T65 demonstrated the system's potential for genetic screening applications. FOT of both cultivars exhibited a significant negative correlation with daily mean temperature. Notably, a temperature-sensitive period was identified in the morning, suggesting that temperature influences not only flower opening but also preceding physiological processes such as panicle and spikelet development. This study presents a novel approach to investigating FOT dynamics in rice and provides insights into the interaction between environmental factors and internal regulatory mechanisms governing this critical reproductive trait.