Research On The Detection Of Torreya Grandis Kernels Using Near Infrared Spectroscopy

Torreya grandis kernels(T.grandis kernels)is a type of high-quality nut with medical and food value,mainly distributed in Kuaiji Mountains and Tianmu Mountains in Zhe Jiang Province.With the rapid development of the T.grandis kernels market,it is necessary to establish a rapid detection method for the T.grandis kernels quality and storage time.This paper proposes a method combining near-infrared spectroscopy with chemometrics to improve the detection accuracy and efficiency of several important parameters for T.grandis kernels quality and storage time.The main conclusions are as follows:(1)The effect of water content on the spectrum was studied and a quantitative analysis model for T.grandis kernels water content was established.The results showed that the greater the water content,the higher the overall absorbance intensity.The water content caused the absorption peak to shift due to the hydration of other components.The R~2c and R~2cv of the PLS quantitative model established after SNV preprocessing procedure were both above 0.97.The model was validated by the regression method,and the corresponding predicted correlation coefficient R_Pwas 0.9635,indicating that the model had high prediction accuracy and good reliability,which could be applied to the accurate detection of water content in T.grandis kernels.(2)To explore the feasibility of non-destructive detection of T.grandis kernels,the protein and fat content of 124 T.grandis kernels samples and the corresponding near-infrared spectra in the shell,shelled and kernel states were measured to explore the best prediction model.For the T.grandis kernels with shell,the protein and fat PLS models had poor predictive performance and did not achieve quantitative prediction.For the T.grandis kernels without shell,the predictive ability of the protein and fat PLS quantitative models was improved,which could be applied to occasions that the sample damage and was small accuracy requirements was low.For the T.grandis kernel,the R~2c and R~2cv of the protein and fat models,and the R_Pof regression model were all greater than0.85,achieving a good detection accuracy of protein and fat content.(3)To establish a classification model for different storage time,near-infrared spectroscopy data for T.grandis kernels with storage time of 4-9 months were collected,and three different classification methods were applied.The results showed that the PCA method could only achieve different between 4-5 months,6 months,7 months,and 8-9months;the training,validation,and prediction accuracy of the SVM method were 98.93%,97.95%,and 97.33%,respectively,and the with poor classification performance of kernels storage for 8 and 9 months;the LDA method had the best classification accuracy for the correction and validation set,which were 100.00%and 97.33%,respectively,achieving a proper classification of Torrey with different storage time.The quantitative models for important quality parameters of T.grandis kernels and qualitative analysis models for T.grandis kernels discrimination with different storage times were developed to provid a theoretical basis for the rapid detection and determination of T.grandis kernels quality parameters and storage time.