Prediction of rice protein content in cold region based on leaf spectral reflectance

A real-time prediction of rice protein content can be expected to realize using the rice leaf spectral index. In this study, the spectral reflectance was collected from the first leaf (L1), the second leaf (L2) and the third leaf (L3) at the top of the main growth period (T1: joining stage, T2: heading stage, T3: fruiting stage) of rice in the cold region under different years, nitrogen fertilizers, and varieties. A correlation model was established to clarify the relationship between the spectral index and rice protein content, in order to investigate the spectral reflectance of leaves under different treatments. The accuracy of the model was verified by the P-k, Root Mean Square Error (RMSE), and Symmetric Mean Absolute Percentage Error (SMAPE). The results showed that the protein content of rice was significantly affected by the nitrogen application rate and variety difference. In the nitrogen fertilizer test, the more nitrogen applied, the higher the protein content of rice was. The protein content values of A8 level were 34.55%, 27.44%, 26.39%, 22.19%, 18.07%, 14.39%, and 12.23% higher than those of A1, A2, A3, A4, A5, A6, and A7, respectively. The taste value of the A8 level decreased by 8.10%, 5.06%, 4.99%, 4.10%, 3.45%, 2.96%, and 2.28%, respectively, compared with the A1-A7 groups. Furthermore, the protein content was ranked in the descending order of the C6>C5>C2>C3>C4>C1 in the variety test, whereas, the taste value was in the order of C1>C4>C3> C2>C5>C6. There was a similar change trend in rice protein content and taste value score under different treatments in 2021 and 2020. A negative correlation between protein content and taste value was found, where the R2 value was 0.93, and the fitting equation was Y=-4.21X+113.32 (Y was the rice taste value score, X was the rice protein content). There was a significant effect of nitrogen application rate on the reflectance of rice leaves, which increased the input of nitrogen fertilizer. In the three periods, the reflectance of rice leaves decreased in the visible region under the same wavelength. There was more outstanding at the “green peak” of reflection at 560 nm to the "red valley" of absorption at about 685 nm. The reflectance of the blade increased in the near-infrared platform. Much more outstanding intensity was found at the band of 760-1 000 nm “reflecting platform”. In the variety test, the spectral reflectance in the visible region was ranked in the ascending order of the L1T3>T1. This variation was closely related to the nutrient transport and protein synthesis in rice. A correlation analysis was performed on the spectral index and rice protein content. An excellent fitting relationship was found with the protein content in the ARI1 ((1/R550)-(1/R700)) of L1 leaf at T2, CTR1 ((R695/R420) of L2 leaf, and Rg (the maximum band reflectance within 510-560 nm of green light) of L3 leaf. Specifically, the T3 stage R2 values were 0.57, 0.75, and 0.63, the P-k values were 0.01, 0.01, and 0.03, the RMSE values were 0.19, 0.11, and 0.14, the SMAPE values were 1.56%, 1.24%, and 1.44%, respectively. Among them, the L2 leaf CTR1 performed best in the T2 period, where the fitting equation with the protein content (Y was protein content, X was index value) was Y=1.24X+2.92, with the R2 value of 0.75. In conclusion, the fast, non-destructive, and real-time prediction of rice protein content was achieved with the help of the CTR1 index, fully meeting the requirements of quality harvest and real-time quality monitoring. The finding can provide a strong reference to promote the sustainable development of high-quality rice in cold areas.