Research And Experiment On Classification Of Fresh Tea Leaves Based On Near-infrared Spectroscopy

There are single fresh leaves with uneven age and tenderness in the raw materials of machine-picked tea.These fresh leaves are often directly processed into bulk tea without grading,so that the machine-picked tea fresh leaves with high tenderness miss the opportunity to be processed into high-quality melon sliced tea.The economic benefits of tea have not been fully utilized.In view of the above problems,this paper uses near-infrared spectroscopy(NIRS)and machine learning technology to grade single fresh leaves with different tenderness grades,including grade calibration of fresh tea leaf samples,spectral data collection,data preprocessing,and feature extraction.Established a grading model for fresh tea leaves,and designed static,dynamic and applicability tests to verify the grading method studied.The main research contents of this paper are as follows:(1)Preparation of fresh tea leaves and spectral data collection.Collect the same batch of fresh tea leaf samples,and formulate three grades of test materials: A,B and C according to the characteristics of the fresh leaves.After introducing the NIR spectral analysis principle and spectrometer required to collect spectral data,the spectral data of the front of fresh leaves was collected and analyzed.(2)Preprocessing and feature extraction of fresh tea leaves spectral data.The Moving Average Smoothing(MAS)method and the SG-convolution smoothing method with different window sizes were selected for the original spectrum of fresh tea leaves to carry out smoothing processing,and the Standard Normal Variate(SNV)method and the Multivariate Scattering Correction(MSC)method were selected for deviation.As well as the SNV method combined with the Detrend processing method(SNV+DT)was selected for baseline correction of the original spectrum;select three methods of MAS+SNV,MAS+MSC and MAS+SNV+MSC with a window size of 11 after combined preprocessing as a preprocessing method for spectral data.Principal Component Analysis(PCA)and Successive Projections Algorithm(SPA)were selected to extract features from the three preprocessed spectral data,and the PCA method that retained the first five principal components was selected to extract features from the spectral data.(3)Research on the classification modeling method of fresh tea leaves.Establish Support Vector Machine(SVM),Extreme Learning Machine(ELM)and K-Nearest Neighbor(KNN)models to classify and identify three grades of fresh tea leaves,and use the model training set,test set accuracy and model confusion matrix as The evaluation index of the model classification effect.Exploring the effects of three sample division methods: Random Selection(RS),Kennard-Selection(K-S)method and Sample Set Partitioning Based on Joint X-Y Distances(SPXY)method on the classification ability of the three models.After experiments,it is concluded that the best classification models based on the three algorithms are MAS+SNV+RBF-SVM,MAS+SNV+ELM(number of hidden layer neurons: 50)and MAS+SNV+F-KNN(k=3),the optimal sample division algorithm is SPXY method.(4)Test and verification of the classification method of fresh tea leaves.Firstly,a static classification verification test of fresh tea leaves was designed to compare the classification performance of SVM,ELM and KNN models in different sample sets.The SVM model has the best recognition effect,with accuracy,true rate,recall rate and harmonic mean The comprehensive average of 96.908%.The tea fresh leaf grading test bench was designed and debugged from the three aspects of hardware,software and control system,and a dynamic verification test was designed to grade 40 fresh leaves of grade A,B and C.The average accuracy of the two tests was 85.417%.Designing A research experiment on the applicability of the fresh tea leaf classification method.And the results show that the average accuracy rates of the fresh leaf models in different seasons and varieties are 92.777% and 95.833%,indicating that the classification method has better effects on fresh leaves of different seasons and varieties.It provides technical support and development ideas for the classification of single fresh leaves in machine-picked tea using NIRS technology.