Rapid Nondestructive Testing Of Rice Based On Hyperspectral Imaging Technology

Rice is one of the food crops with high nutritional value and can provide a large amount of energy for the human body.About 50%of the world’s population uses rice as their main food source.In China,rice production accounts for 67%of the world’s total.Moisture,starch and protein are three important chemical substances in rice,and measuring their content can evaluate the quality of rice.Hyperspectral imaging technology is a research hotspot in the field of non-destructive testing,and its advantages are that it can combine spectra with images,and realize online detection of samples and analysis of chemical composition changes.By using hyperspectral imaging technology to determine the chemical composition of rice,a predictive model for the rapid detection of starch,protein,and moisture in rice was established.By further simplifying the prediction model,the experimental cost was reduced,and the chemical composition content distribution was visualized.1. By measuring the chemical composition of rice,the water content of 87different varieties of rice samples was distributed between 9.8000% and 16.6000%;the average starch content of 87 different varieties of rice was between 42.8000%and65.3000%;87 different varieties The average protein content of rice is between5.1000%and 9.3000%.Lianjing 11,which is a national standard third-grade high-quality rice,is superior in the main nutritional value level.2. By scanning rice samples with hyperspectral imaging equipment,the obtained spectral image range is 1000-2500 nm.Spectral images show the same trend at wavelengths from 938 nm to 2215 nm;at wavelengths below 938 nm and above 2215nm,the curve shows an irregular trend.Due to the strong radiation and noise,the image is cluttered.3. A hyperspectral scan image was combined with partial least squares regression(PLSR),principal component analysis regression(PCR),and support vector regression(SVR)to establish a full-spectrum calibration model of rice starch content.The calibration correlation coefficient R_C~2 of the PLSR model was obtained as0.9040.Based on three multivariate statistical methods,a full-spectrum prediction model of rice was established according to the established correction model.Among them,the prediction correlation coefficient of the full-spectrum PLSR prediction model of rice hyperspectral starch content was R_P~2=0.8147 and the predicted mean square error RMSEP=0.0245.The PLSR starch prediction model has basic prediction capabilities.In order to simplify the stoichiometry model,the characteristic wavelength of rice starch was selected based on the correlation coefficient of starch PLS,and the characteristic wavelengths were determined to be 1051,1158,1346,1720,1809,and 2180nm.Based on 7 characteristic wavelengths,5 PLSR models were established for starch prediction.The modeling effect is compared,and the 7-wavelength model is determined to be a simplified model for predicting the optimal PLSR starch content,which has a high prediction correlation coefficient.By converting the 7-wavelength partial least squares model to each pixel of the image to predict the rice starch content,a visual map of rice starch content is constructed,which can more directly reflect the rice starch content through color changes.4. A full-spectrum calibration model of rice protein was established using PLSR,PCR,and least squares support vector(LS-SVR)regression with hyperspectral scan images.The calibration correlation coefficient R_C~2 of the PLSR model was 0.8894.Based on three multivariate statistical methods,a full-spectrum prediction model of rice protein was established according to the established correction model.Among them,the prediction correlation coefficient of full-spectrum partial spectrum PLSR prediction model of rice hyperspectral protein content was R_P~2=0.8434.The PLSR protein prediction model has basic prediction capabilities.In order to simplify the stoichiometric prediction model,the characteristic wavelength of the protein was selected.Compared to moisture and starch,protein characteristic wavelengths present a more complex distribution.According to the significant PLS regression coefficients,1026,1152,1233,1290,1402,1471,1648,1744,1792,1855,1930,1970,2036,2069,2101,2138,2164 nm were used as significant regression coefficients in this study Characteristic wavelength.A simplified model for predicting protein content in PLSR was established based on characteristic wavelengths.A simplified model for17-wavelength prediction and a simplified model for 7-protein typical wavelength were established.Finally,the 17-wavelength model was used as the final simplified model for the prediction of rice hyperspectral protein content.By converting the above-mentioned 17-wavelength PLSR model to each pixel of the image to predict the protein content of rice,the rice starch content distribution of 8 different varieties in the prediction set is converted into a visual image,which can more intuitively understand the protein content distribution in rice.5. A full-spectrum calibration model of rice moisture was established by using hyperspectral scan images combined with PLSR.The calibration correlation coefficient R_C~2 of the PLSR model was 0.8588.According to the statistical method of PLSR multivariate data,the full-spectrum of rice moisture was established according to the established calibration model Prediction model,in which the prediction correlation coefficient R_P~2=0.8159 of the full-spectrum PLSR prediction model of rice hyperspectral moisture content.The PLSR moisture prediction model has basic prediction capabilities.In order to simplify the stoichiometric prediction model,a simplified model of the characteristic wavelength of moisture was established.Based on the significant PLS regression coefficients,1158,1214,1321,1427,1732,1867,2112,and 2170nm were used as characteristic wavelengths with significant regression coefficients in this study.Based on the characteristic wavelength,a simplified model for predicting moisture content in PLSR was established,and a simplified model for predicting 8-wavelength was established.Finally,the 8-wavelength model was used as the final simplified model for predicting hyperspectral moisture content in rice.The rapid and non-destructive detection of rice components by hyperspectral imaging technology can deepen the understanding of rice component content distribution and have important significance for rice agricultural production and harvesting processes.