Objective A simple and rapid method to accurately estimate the nitrogen content in grains of a rice plant was developed based on features derived from photographic images.

Methods A field experimentation applying nitrogen fertilizations at 0 kg·hm⁻² (N0), 75 kg·hm⁻² (N1), 150 kg·hm⁻² (N2), and 225 kg·hm⁻² (N3) on the lots of growing 4 cultivars including the local major rice Zhongjiazao 17 and hybrid Changliangyu 173 to take pictures was conducted from 2022-2023. A Canon EOS 100D digital camera with a resolution of 72 DPI was used to obtain images of canopies of the growing rice plants. Grain biomass and nitrogen content were conventionally measured simultaneously to correlate with various derivative features of the images to establish a mathematic model of predicting nitrogen content of the rice plants.

Results The highest correlation coefficients of the regression equations were found between the percentage of rice pixels (PRP) and features derived from the images and the leaf area index (LAI), above ground biomass (AGB), and nutrition accumulation (PNA) of the plants. The prediction model for rice at jointing stage rendered the most accurate estimations. The coefficients R² of the single PRP-based models on LAI, AGB, and PNA were 0.86, 0.76, and 0.52, respectively (P＜0.01), and the RMSEs for the model validation 0.32 g·m⁻², 22.30 g·m⁻², and 2.54 g·m⁻² at the RRMSEs of 8.25%, 7.61%, and 26.49%, respectively. Whereas the high-order exponential function of the PRP derivatives predicted LAI, AGB, and PNA, with R² of 0.96, 0.99, and 0.94, respectively (P＜0.01) at the RMSEs of 0.16 g·m⁻², 3.71 g·m⁻², 0.57 g·m⁻², and RRMSEs below 10% at 4.20%, 1.27%, and 5.98%, respectively, indicating a significant high stability of the prediction model.

Conclusion It appeared that accurate estimation of grain nitrogen content of a rice plant could be obtained by using the features derived from photographic images of the canopy.