A Method Of Tea Quality Inspection Based On Hyperspectral Imagine

The internal quality of the tea plays a decisive role in the taste and aroma of the tea.The commonly used method for identifying the internal quality of the tea is to carry out physical and chemical measurements.This method is tedious,time-consuming and expensive,and it damages the test sample and seriously wastes resources.Therefore,it is of great significance to study a convenient,fast and non-destructive internal quality inspection method.This article mainly detects the three components in tea-water,tea polyphenols,and free amino acids.The experiment selected one bud,one bud one leaf,one bud two-leaf tea fresh leaves and four grades of Ya’an Ganlu green tea as experimental samples.The experimental data mainly includes two parts,one is to obtain the spectral data of the sample based on the hyperspectral imaging system,which mainly includes the preprocessing of the hyperspectral data and the extraction of the spectral data of the region of interest;the second is the determination of the physical and chemical values of the corresponding components,mainly based on the current national standards Determine the physical and chemical values of the ingredients.Based on the spectrum-physical and chemical data,a quantitative detection model for the moisture content of fresh tea leaves,a quantitative detection model for the total amount of polyphenols in nectar green tea and a quantitative detection model for the total amount of free amino acids in nectar green tea are established.SG convolution smoothing（SG-Smoothing）combined with multivariate scattering correction（MSC）was used to pre-treat the spectrum in the tea moisture content detection model.Using non-information elimination method combined with continuous projection algorithm（UVE-SPA）to extract features from the pre-processed spectrum,a total of 23feature bands were extracted.A support vector regression algorithm was used to construct a quantitative detection model for the moisture content of fresh tea leaves,and a grid search algorithm was introduced to optimize SVMR.Finally,the optimal UVE-SPA-SVMR quantitative measurement model of fresh tea leaf moisture content was constructed.The mean square error and determination coefficient of the training set were RMSEC=0.01236 and R_c²=0.9031;the mean square error and determination coefficient of the test set were respectively RMSEP=0.0241 and R_p²=0.8764.SG convolution smoothing（SG-Smoothing）combined with standard normal variable transformation（SNV）combined with 4th order detrend(Detrend-4^st)was used to pretreat the spectral data in the quantitative detection model of polyphenol content in nectar green tea.Joint interval partial least squares（siPLS）and competitive adaptive reweighting algorithm（CARS）are used to extract the feature spectrum of the preprocessed spectrum,and finally extract 10 feature bands to form the optimal spectral data set.An error back propagation neural network（BPNN）was used to construct a quantitative detection model of tea polyphenols.The mean square error and determination coefficient of the training set were RMSEC=0.0050 and R_c²=0.9444;the mean square error and determination coefficient of the test set were respectively RMSEP=0.0094 and R_p²=0.9344.In the quantitative detection model of the content of free amino acids in nectar green tea,multi-spectral scattering correction（MSC）spectral data was used for pretreatment.Continuous projection algorithm combined with kernel principal component（SPA-KPCA）was used to extract feature spectrum,and finally obtained 3 feature information to constitute the optimal spectral data set.An extreme learning machine（ELM）was used to construct a quantitative detection model of free amino acids in manna green tea.The mean square error and determination coefficient of the training set were RMSEC=0.0504 and R_c²=0.9343;the mean square error and determination coefficient of the test set were RMSEP=0.0700 and R_p²=0.9170.