Quality Analysis And Category Identification Of Rice Based On The Near Infrared Spectroscopy (NIRS)

Near infrared reflectance spectra (NIRS) is a kind of electromagnetic wave between visible spectrum and middle infrared spectrum. Recently, NIRS has been developed as a kind of novel analysis and research method due to its characteristics of fast, convenient and green environmental protection.With the general improvement of living standards, people pay more and more attention to the quality of rice. Except for sensory analysis external features, such as color, shape, smell and others. In order to determine the merits of eating quality of rice, we analyze the mass fractions of water, protein and amylose in rice by chemical methods, and identify rice varieties mainly through the sensory evaluation and the experience. As the chemical detection method is cumbersome, time-consuming and sensory analysis is interferenced by subjective factors, the development of a rapid quality inspection and varieties identification method for rice is extremely urgent. NIRS is a newly developed method for analysis and research. Depend on the acquired complete spectrum information of rice samples, construct mathematical models by mathematical methods which can fully represent the characteristics of whole spectrum and exclusive attributes of rice samples. It would be able to find out the exclusive attributes of rice samples covered by lots of complex common factors. What is more, it would possibly be able to depict the resemblances among spectrums form different species of rice samples quantitatively or qualitatively. That is the reason why near infrared reflectance spectra (NIRS) can be used for rice quality evaluation and category analysis.In this thesis,102rice samples were employed by means of NIRS analysis techniques. Based on the chemical analysis data of3major components for the102rice samples by national approved standard methods, the quantitative analysis models were developed by the method of NIRS. During the process of model developing,11kinds of spectrum pretreatment methods are employed, including constant offset elimination, subtract a straight line, vector normalization, min-max normalization, multiplication scattering correction， first derivative, second derivative, first derivative+subtract a straight line, first derivative+vector normalization, first derivative+multiplication scattering correction and no spectral data preprocessing. Eighty percent of samples were picked at random to construct models, then the left twenty of samples were used for models externally test. Quantitative analysis models of rice moisture, protein and amylose were developed by partial least square (PLS). The accuracy of the prediction result was evaluated. Internal cross-validation decided coefficient (R2) of prediction model was0.992,0.9792and0.9736, respectively. Internal cross-validation RMSECV was0.141,0.201and0.209, respectively. External validation determination coefficient (R2) of model was0.9861,0.912and0.9373, respectively. External validation RMSEP was0.179,0.206and0.243, respectively. While use standard method the differences among samples could be tested by calculation of the euclidean distance between near infrared reflectance spectra of samples. The differences of category of rice samples were evaluated by cluster analysis. Results showed that NIRS, a rapid and non-destructive analytical technique, can be used for rice quality evaluation and category analysis.