Study On The Rice Quality Analysis Model Based On Near Infrared Technology

Rice is one of the main food crops in our country,it is difficult to evaluate quality due to wide range of rice varieties,researchers can not discriminate rice quality rely on appearance,size of rice,the traditional rice quality analysis method is not suitable for field acquisition.We try to make research of models for rapid analysis of rice moisture content,crude protein content,fatty acid value and amylose content base on near infrared spectrum technology.All of this work was designed to provide a fast and accurate method for rice quality analysis in field.In the experiment,conventional chemical methods were used to measure chemical values of rice samples,such as moisture content,crude protein content fatty acid value,amylose content.The distribution range of measured moisture content,crude protein content and fatty acid value,amylose content were 10.26 ~21.26%,10.26~ 21.26%,11.44 ~ 41.89 mgKOH / 100 g,41.89 ~ 20.86%,respectively.SupNIR-2700 series of near infrared spectrum analyzer was used to collect rice samples spectrum of Anhui province,Jiangsu province,Hunan province and Heilongjiang province.Taking chemical metrology calibration software RIMP for data analysis which attached to the instrument and choosing mahalanobis distance method to delete abnormal samples of calibration set samples,comparing modeling effect of different preprocessing methods by the SEC,RC,SECV,SEP four indicators,selecting the optimal preprocessing method for PLS and ANN.By adopting the random method to divide 80% samples as calibration set and another 20% as validation set,choosing the partial least squares(PLS)and artificial neural network(ANN)to establish the near infrared quantitative analysis model of rice quality indicators including japonica rice,indica rice and all rice samples calibration model,comparing the difference between PLS and ANN.Appropriate numbers of unknown samples were selected to analysis predictive ability of calibration models,two-tailed T test was chosen to analyze whether there was a significant difference between predicted value and chemical value to explore near infrared spectrum technology can be used for accurate determination of rice main quality indicators or not.Results showed that :1.There were some differences in modeling results under different pretreatment methods.For PLS modeling,preprocessing method of Savitzky-Golay smooth +Savitzky-Golay derivative + multiple scatter correction(MSC)+ average centralized had optimal effect,for ANN modeling,preprocessing method of standard normal variable +transformation trend correction + Savitzky-Golay smooth + Savitzky-Golay derivative + average centralized had optimal effect.2.There were some differences in modeling results among different regions and different kinds of rice.There were no obvious difference in modeling between PLS and ANN,they can both be used for near infrared models building of rice main quality indexes.3.For unknown samples,the model of rice moisture content determination coefficient(R2)of prediction was close to 0.97,absolute deviation was 0.06%between chemical values and predicted values,the model of rice crude protein content determination coefficient(R2)of prediction was close to 0.87,absolute deviation was0.02% between chemical values and predicted values,the model of fatty acid value determination coefficient(R2)was more than 0.62,absolute deviation was 0.13 mg KOH / 100 g between chemical values and predicted values,the model of rice amylose content determination coefficient(R2)was more than 0.84,absolute deviation was0.08% between chemical values and predicted values,two-tailed t-test results showed that there was no significant difference between chemical values and predicted values.4.Near infrared spectrum technology can be used for rapid detection of rice moisture content,crude protein content,fatty acid value and amylose content.The model had accurate prediction of rice moisture content,crude protein content and amylose content,and the prediction of fatty acid value was well.