Abstract:
Abstract: A quantitative estimation of nitrogen content in rice leaves has been one of the most critical steps to realize precision fertilization and variety breeding in modern agriculture. Optical remote sensing technology has been emerging in recent years. The multi- and hyper-spectral remote sensing data can be utilized to rapidly obtain the physical and chemical parameters, such as the rice nutritional status, pest stress, and phenotypic information. Therefore, it is very necessary to implement the high-throughput acquisition of breeding phenotypes in the rice digital production using spectral technology. In this study, a rapid and accurate inversion of nitrogen content was realized to detect the spectral characteristics and variation in the rice leaves using the character transfer vegetation index in the spectral technology. The results showed that there were all the same changes in the spectral curves of rice leaves in the key growth periods, indicating the extremely significant characteristics of "two valleys and one peak" in each period. The spectral characteristic curve varied significantly in the visible and near-infrared regions in the period of rice growth. Specifically, the spectral curve changed greatly in the near-infrared region from the tillering to the heading stage, whereas, the hyperspectral reflectance increased gradually in each growth period. The characteristic bands of the rice nitrogen content were extracted from the continuous hyperspectral reflectance with lots of redundant information. The vegetation index of nitrogen content inversion was then constructed for the rice leaves. Since the continuous projection was a forward band selection, the projection of each cycle was calculated in the remaining band starting from a band variable, and then the band corresponding to the maximum value of the projection vector was introduced into the band combination, while ensuring the lowest correlation between the selected and previous band, finally to repeat the above steps until the selected number of bands were fully met the given requirements. Therefore, the continuous projection was used to screen the characteristic bands of rice hyperspectral in the band of 400-1000 nm, where the correction set was utilized to conduct the internal cross validation of the selected bands. Six hyperspectral characteristic bands of nitrogen content in rice leaves were screened, i.e., 500, 555, 662, 690, 729, and 800 nm, according to the Root Mean Square Error of Cross Validation (RMSECV) values in the validation. Once the form of vegetation index was determined, the current vegetation index of rice nitrogen content was mostly constructed to change the bands for a new one. The mature form of vegetation index was normally used to determine the optimal band, according to the different inversion parameters. Nevertheless, there was still lacking of relationship between nitrogen content and spectral characteristics. Therefore, the multiple characteristic bands were first converted into three bands for the nitrogen characteristic transfer index, according to the band characteristic transfer. A linear regression model was established to verify the inversion model of nitrogen content in rice leaves. Taking Nitrogen Characteristic Transfer Index (NCTI) as the input, the inversion model of rice nitrogen content was constructed by linear regression. The determination coefficient of the model was 0.774, and the root mean square error was 0.379 mg/g. The inversion performed better, compared with the traditional vegetation indexes, such as NDVI (Normalized Difference Vegetation Index) and EVI (Ratio Vegetation Index). Consequently, the NCTI can be widely expected to serve as a hyperspectral vegetation index for the rapid inversion of nitrogen content in rice leaves in practical application. The finding can provide promising data support and strong reference for the spectral detection of nitrogen content in rice leaves.