Study On Tea Variety Classification And Adulteration Degree Detection Based On Hyperspectral Image Technology

In the process of tea sales,some illegal traders often use cheap tea to fake expensive tea or mix expensive tea with other similar tea to pursuit huge profits,and it is difficult for ordinary consumers to judge the actual tea varieties only by their naked eyes.Traditional methods for tea variety identification include artificial sensory discrimination and chemical analysis methods.However,artificial sensory discrimination lacks objectivity because it relies on subjective experience.Although chemical analysis methods are accurate and reliable,they will cause damage to samples and cost lots of time and effort.In recent years,spectroscopy has been widely used in the discrimination of tea varieties.Spectroscopy uses point light source to sampling,which makes the detection performance easy to be influenced by sampling location.By sampling the whole sample,hyperspectral image technology can collect more comprehensive information of sample,which can effectively make up for the shortcomings of the spectroscopy.In this paper,five varieties of oolong tea and Tieguanyin tea adulteration degrees were detected based on hyperspectral image technology,and the system for tea variety classification and adulteration degree detection was designed.The main content and conclusions of this paper are as follows:(1)Five kinds of oolong tea(Tieguanyin,Jinguanyin,Benshan,Huangjingui,Maoxie)were distinguished by hyperspectral image technology.Firstly,hyperspectral images of five kinds of oolong tea were obtained by hyperspectral imaging system,the original spectral data was extracted from hyperspectral images.Secondly,the redundant wavelengths at the beginning and end were removed,and the spectral data was preprocessed by multiplicative scatter correction(MSC).Thirdly,the characteristic wavelengths were selected by competitive adaptive reweighed sampling(CARS)and bootstrapping soft shrinkage(BOSS),respectively.Fourthly,extreme gradient boosting(XGBoost)and light gradient boosting machine(LightGBM)models based on the above characteristic data were established respectively,and grid search(GS)method was used to optimize the model parameters.Eventually,the experimental results showed that BOSS-GS-LightGBM model had the best classification effect,with accuracy of 100% for training set and 97.33% for test set.(2)The adulteration degrees of Tieguanyin tea were detected based on hyperspectral image technology.Comparing the economic value gap,appearance difference and data separability among different tea varieties,Benshan tea was chosen as the "adulterant" of Tieguanyin tea in this paper.Tieguanyin tea and Benshan tea were mixed at different proportions and hyperspectral images of adulterated samples were collected.After extracting the spectral data from hyperspectral images,redundant spectral data at the beginning and end were removed.Then,MSC algorithm was used to preprocess the extracted spectral data,and the characteristic wavelengths were selected by iteratively retaining informative variables(IRIV)and BOSS algorithm,respectively.Moreover,GS algorithm and genetic algorithm(GA)were used to optimize the parameters of XGBoost and LightGBM to obtain the best model parameters,respectively.The final results showed that BOSS-GA-LightGBM model had the best classification effect,with accuracy of 100% for training set and 91.82% for test set.Therefore,it is feasible to use hyperspectral image technology and BOSS-GA-LightGBM model to classify Tieguanyin tea with different adulteration degrees.(3)The system of tea variety classification and adulteration degree detection was designed.The system included two parts: oolong tea variety classification and Tieguanyin tea adulteration degree detection,which could realize the functions of importing,processing,querying and exporting data.After completing the design of the system,20 tea samples with different varieties and 33 adulterated tea samples with different adulteration degrees were used to verify the detection effect of this system,and the accuracies were 95% and 90.91% respectively,indicating that this system can effectively detect different oolong tea varieties and Tieguanyin tea adulteration degree.In this paper,hyperspectral image technology was used to distinguish different varieties of oolong tea and different adulteration degrees of Tieguanyin tea,and finally the relevant detection system was designed.The above research not only provides a theoretical basis for the tea variety classification and adulteration degree detection,but also lays a foundation for the development of accurate and nondestructive detection instrument.