Estimating rice flower intensity using flower spectral information from unmanned aerial vehicle (UAV) hyperspectral images

Growth monitoring of rice is of great significance to food security of human society. Rice flowering is an important growing stage for grain formation, and flower intensity is the dominant factor in determining rice yield. The estimation of flower intensity helps us to know the rice yield in advance. This research proposed a series of the flower index (FI) to monitor status of rice flowers by developing flower intensity estimation models using stepwise multiple linear regression (SMLR) and random forest (RF), and their performance was compared to the models developed by vegetation indices (VI) of some key growth stages. The FI that in types of normalization (NDFI), ratio (RFI) and differences (DFI) were tested. The involved FIs in the FI-based models were those in type of difference (DFI) that obtained by difference of reflectance before and after flowering (DR) and their first derivative (DR'). The FIs of ten consecutive days during the flowering were obtained and their correlations with flower intensity showed that FIs of the late flowering (the 8th day, 9th day and 10th day) were more significantly correlated to flower intensity than those at the early or mid-flowering, with the maximum correlation coefficient of 0.702 given by FIs in difference type formed by DR'. The accuracy assessment of flower intensity estimation models showed that FI-based models had the equivalent accuracies to VI-based models, especially for the SMLR model that based on four FIs, which had R2 = 0.707, MAPE =10.54%, rRMSE =11.39%, was comparable to the model that developed by three VIs of the booting, heading and jointing stages (R2 = 0.751, MAPE = 8.99%, rRMSE = 10.31%). The promising results of FIs in estimating flower intensity make the simple data acquisition possible and provide an alternative way to get information about rice yield.

Automated summary

Monitoring the growth of rice is crucial for ensuring food security. This research developed flower intensity estimation models using stepwise multiple linear regression (SMLR) and random forest (RF) to monitor the status of rice flowers. The results showed that the accuracy of flower intensity estimation models based on flower indices (FI) was comparable to models based on vegetation indices (VI). These promising results provide an alternative way to obtain information about rice yield.