Study On Classification Of Tea Levels Based On Electronic Nose And Near Infrared Spectroscopy

Our country’s tea culture is profound.Tea is not only an economic crop,but also gradually becomes one of our country’s core cultural competitiveness.The health effects of tea have gradually received various scientific demonstrations.As people gradually deepened the concept that tea has health effects,the output of tea has increased year by year.At present,the standardization of tea grades is still very vague,and the realization of accurate identification of tea producing areas,varieties and grades has always been a hot topic discussed by scholars.In order to solve the above problems,this research establishes a tea grade classification model,realizes a more accurate classification of tea grades,based on electronic nose technology and near-infrared spectroscopy technology and combined with a series of bionic intelligent algorithms and information fusion technology,which will promote standardization of tea grade classification.(1)First of all,this research uses an electronic nose instrument and a near-infrared spectrometer to detect and record the data of tea samples.Based on data analysis,we normalize the original data and use convolution filtering(Savitzky-Golay,SG)to preprocess the original spectrum.Then,principal component analysis(PCA)is used to reduce the dimensionality of the normalized tea data and reduce the redundant data to select the first several features.Based on electronic nose technology and near-infrared technology,using support vector machine(SVM)as the basic classification model to classify the grades of Dianhong tea,the classification accuracy rates are 74.29% and85.71% respectively,which provide a basis for comparison for the subsequent optimization model.(2)Based on the electronic nose technology,the basic particle swarm optimization(PSO)is proposed to optimize the SVM model,in order to realize SVM model parameter optimization.The method is used to search for the optimal combination of penalty factor and kernel function parameter,and the classification accuracy rate of Dianhong tea is 86.43%.In order to further improve the classification accuracy of the model,this paper proposes a comprehensive learning particle swarm optimization algorithm(CLPSO)to optimize the SVM model based on the PSO algorithm.Compared with the basic PSO algorithm,this new algorithm changes its particle learning method.Particles can learn from fine particles in different dimensions.At the same time,inertial weights are introduced to the new algorithm,which strengthens the algorithm’s global and local optimization capabilities.It is effective to control the convergence rate of the algorithm.Finally,a classification accuracy of 87.86% is obtained.(3)Based on the near-infrared technology,this research proposes to use the differential evolution algorithm DE1(DE/rand/1/bin)and DE2(DE/current-tobest/2/bin)to optimize the SVM classification model.The difference between DE1 and DE2 is that the number of difference vectors and the number of scaling factors required in the mutation process are different,while both algorithms can prevent the algorithm from falling into local optimality or algorithm stagnation.The DE1-SVM model and DE2-SVM model can obtain 95.0% and 97.86% classification accuracy for Dianhong tea grades,respectively,but for Huangshan Maofeng,Meicha and Qihong tea,even the DE2-SVM model can only get 90%,93.3%,92.14% accuracy rate.This shows that the use of DE algorithm to optimize the SVM model can significantly improve the classification accuracy of certain types of tea(such as Dianhong tea),but for other teas,the improvement is relatively limited.(4)Based on the electronic nose technology and the near-infrared spectroscopy technology,this research also uses the information fusion technology to establish the information connection between the electronic nose and the near-infrared,amplify the detailed information of the sample,and weaken the irrelevant information.Based on the CLPSO algorithm and the BBO algorithm,the Geographical Particle Swarm Optimization(BLPSO)algorithm optimizes the SVM model,which combines comprehensive learning strategies and biogeographical learning strategies to achieve a more accurate classification of Huangshan Maofeng,Meicha,Dianhong and Qihong tea grades.After analyzing the electronic nose data and near-infrared data and preprocessing them,we use the method of information fusion on the feature vector of the electronic nose and near-infrared data in the feature layer to obtain a new feature matrix as the new model input.Finally,we obtain classification accuracy of the four tea grades of Huangshan Maofeng,Meicha,Qihong and Dianhong are 90.83%,97.5%,93.57% and 98.6% respectively.