Research On Tea Quality Detection Method Based On Spectroscopic Technology

Tea culture is an important cultural heritage in China,and the tea industry is a distinctive and advantageous industry that can promote economic development and promote national culture.However,the wide variety of tea types and the complex grading system have led to dishonest practices such as passing off inferior products as high quality,which has had a negative impact on the tea industry.In order to promote the healthy development of the tea industry,it is necessary not only to strengthen market supervision and management to curb malicious behavior by some merchants,but also to use technical means to provide objective and fair testing results of tea quality level,thereby helping consumers to improve their ability to identify tea quality.In this regard,this article proposes a fast and non-destructive method for detecting tea quality level based on Fourier transform near-infrared spectrometry and hyperspectral imaging technology.The specific research content is as follows:(1)A tea quality level detection method based on near-infrared spectroscopy is proposed,which realizes the fast detection of the quality level of a single type of tea.First,the absorbance spectra of six levels of Huangshan Maofeng tea samples were obtained in the wavenumber range of 4000～10000cm-1 using near-infrared spectroscopy.The absorption spectral data was then converted to transmittance data,and the Savitzky-Golay(SG)algorithm was used to smooth the spectral data.Principal component analysis(PCA)was then used to reduce the dimensionality of the noise-removed transmittance spectra data.The support vector machine(SVM)level classification model was established with the PCA-derived feature variables as input variables and the tea quality level as output variables.To improve the classification performance of the support vector machine model,particle swarm optimization(PSO)and comprehensive learning particle swarm optimizer(CLPSO)algorithms were used to find the best penalty factor c and kernel function parameter g.Comparative experiments were also conducted using the traditional classification model,partial least squares-discriminant analysis(PLS-DA).The experimental results showed that the CLPSO-SVM method had the best classification performance,with a classification accuracy being 99.17%,which proved that the application of Fourier transform near-infrared spectrometer technology combined with swarm intelligence optimization algorithms can effectively achieve fast and non-destructive detection of tea quality level for a single type of tea.(2)Based on hyperspectral imaging technology and advanced deep learning technology,a fast detection method for multiple tea quality level was developed.Firstly,hyperspectral images of seven levels of Keemun black tea samples and seven levels of Chunmee green tea samples were collected.Then,14 feature spectral profiles were obtained using the Competitive Adaptive Reweighted Sampling(CARS)algorithm,which can effectively process high-dimensional hyperspectral data.Next,Principal Component Analysis(PCA)was used to select 8 final profiles from these 14 feature spectral profiles(the cumulative contribution of these 8 feature spectral profiles reached 99.92% of the original hyperspectral image),and Extended Morphological Profile(EMP)was used to augment the data.Finally,a twodimensional mixed convolutional neural network was designed to detect the quality level of the two types of tea based on the processed hyperspectral images.The detection accuracy was 95.05%,providing an effective method for fast and accurate detection of the quality level of multiple types of tea.