Rice Variety Identification And Protein Content Prediction Based On Hyperspectral Imaging

Rice is one of the most important food crops in my country.The safety and quality of its production process and the quality of the harvested rice play an extremely important role in the daily life of the people and the stability and development of the country.With the continuous development of modern society,people put forward higher requirements for the quality of rice,and to obtain high-quality rice needs to start from two aspects of production process and rice quality inspection.This paper takes rice grains and rice plants in the field as the research object,uses hyperspectral imaging technology,image processing technology and mathematical modeling methods to carry out identification of rice grain varieties,prediction of protein content in rice grains,spectral parameters-agronomic parameters-prediction model of rice grain protein content at canopy scale.It aims to provide new ideas and new methods for the non-destructive testing of rice quality,and provide assistance for rice production.The main research contents are as follows:(1)Rice variety identification based on hyperspectral imaging.Different varieties of rice will have differences in taste and nutritional value.Using hyperspectral imaging technology combined with image processing technology to realize the identification of rice varieties.Collect rice grain samples of different varieties,extract the spectral information of rice grains,select 11 characteristic wavelengths using the Successive Projections Algorithm(SPA),and extract the texture parameters of rice grains at the same time,and use the spectrum,image texture features,fusion data to establish a partial least squares discriminant model(PLS-DA)for different species.When the best latent variables(LVs)of spectral data were 11,the minimum RMSE was 0.3860,the accuracy of modeling set was 86.43%,and the prediction set was78.33%.When the texture feature LVs=6,RMSE=0.6220,the accuracy of the modeling set was 71.43%,and the prediction set was 63.33%;PLS-DA model was established by fusing data.When LVs=15,RMSE=0.3026,the accuracy rate of modeling set was 95%,and the prediction set was 88.34%.(2)Prediction of rice grain protein content based on hyperspectral imaging.The Kjeldahl nitrogen analyzer was used to measure the protein content of rice grains,and different mathematical transformations were performed on the spectral information,and the protein content prediction model of rice grains was obtained by combining the spectra with protein content.After Multiplicative scatter correction(MSC)preprocessing,the Multiple linear regression(MLR)model based on the characteristic wavelengths selected by SPA gave the best prediction performance.The R_C~2of the correction set is 0.9393,the R_C~2of the validation set is 0.8998,RMSEV is 0.1725,RPD is 3.16(3)Prediction of rice agronomic parameters and grain protein content based on hyperspectral imaging.The rice canopy spectrum was collected in the field,and the rice SPAD value,leaf nitrogen content,leaf nitrogen accumulation,and grain protein content were measured simultaneously.Extract 14 vegetation indices from the canopy spectral data,analyze the correlation between the vegetation index and rice SPAD value,leaf nitrogen content,leaf nitrogen accumulation and grain protein content,and establish the prediction model of rice agronomic parameters and grain protein content.The best prediction effect of SPAD value is the PCR model based on NDVI(810,680)and RVI(810,680),R_C~2=0.8005,R_V~2=0.6832,RMSEV=1.3283,RPD=1.74.Among the prediction models of rice leaf nitrogen content,the best prediction effect is the PLSR model based on NDVI(810,680),RVI(810,680)and Bm SR705(750,445),R_C~2=0.8032,R_V~2=0.7233,RMSEV=0.1339,RPD=1.90.Among the prediction models of rice leaf nitrogen accumulation,the best prediction effect is the PLSR model based on NDVI(810,680)and RVI(810,680),R_C~2=0.9235,R_V~2=0.8825,RMSEV=0.5395,RPD=2.92.Among the prediction models of rice grain protein content,the best prediction effect is the quadratic function model based on leaf nitrogen accumulation,R_V~2=0.7919,RMSEV=0.3135,RPD=2.21.