Study On Withering Moisture Sensing Model Of Black Tea Based On Spectrum And Image

Withering is the first process of black tea processing,its quality will directly affect the quality of tea.Moisture content is the main index to judge the moderation of withering.However,most tea factories cannot quickly,non-destructively and accurately determine the moisture content of the withered leaves,and also cannot objectively judge whether the withering is uniform.Therefore,a rapid,non-destructive and accurate method to determine the moisture content and uniformity of withering leaves is urgently needed.Based on hyperspectral image technology,this study established a quantitative moisture content prediction model under the sequence of withering.It can realize moisture visualization of withering leaves through this model,hoping to provide a theoretical basis for improving the quality of black tea and promoting the intelligent development of black tea withering processing.The main research contents of this paper were as follows:Spectral information is better than image information in representing moisture content of the withering leaves.Firstly the spectral information and image information of the samples under the sequence of withered leaves were collected respectively.The moisture content of the samples was measured.Finally,the spectral information was preprocessed by standard variable normalization(SNV).After synergy interval partial least squares(Si-PLS)characteristic bands were screened,the model accuracy was the highest.The 9 color parameters(R,G,B,H,S,V,L*,a* and b*)and 6 texture parameters(m,?,r,?,U and e)were extracted from the image information.These 15 characteristic variables were firstly preprocessed with Z-score and PCA,and then establish the PLSR moisture content prediction model.The RPD values of the two models were both greater than 2,indicating that both spectral information and image information could better express the mositure content in the withering leaves.However,the spectral information could better express the mositure content in the withering leaves.2.The prediction model based on the positive(reverse)spectrum of the withered leaf single leaf cannot predict the mositure content on the reverse(positive).However,the pretreatment and the screening of the characteristic bands could greatly reduce the influence on the model.the influence was produced by the interference of the positive leaves and reverse leaves.Based on the hyperspectral image technology,the positive and reverse hyperspectral images of withering leaves were taken respectively.The spectral information was extracted to establish the partial least squares regression(PLSR)moisture content prediction model of the original spectral information of the single leaf and the established prediction model was cross-verified.It can be verified that the performance of the model decreases greatly.When the positive spectrums were pretreated by first derivative(1Der)and the characteristic bands were screened by Shuffled frog leaping algorithm(SFLA),the reverse spectrums were pretreated by multiplicative scatter correction(MSC)and the characteristic bands were screened by competitive adaptive reweighted sampling(CARS),both the performance of the established PLSR moisture content prediction model and the performance of the cross-verified model were significantly improved compared with the model established by the raw spectrum.It indicates that pretreatment and characteristic wavelength screening can reduce the influence of the positive and reverse errors on the moisture content prediction model,but cannot completely eliminate the influence.3.Moisture content prediction model and visualization research based on stacked withering leaves.Aiming at the problem that the spectral errors generated by the positive and reverse of the withering leaves affect the performance of the water content prediction model.Using the stacked withering leaves as the research object to capture the hyperspectral image information and extract the spectral information from the hyperspectral image.And then pretreatment the spectral information to eliminate the problems of baseline drift,high frequency noise and optical path scattering.Through feature wavelength extraction to reduce the redundant information of the input model,partial least squares regression(PLSR)linear model and support vector machine regression(SVR)nonlinear model were established respectively and the performance of the model were compared.Finally,Si-sfla-svr nonlinear model was determined as the optimal moisture content prediction model.The moisture content of each pixel in the hyperspectral image is calculated by this model,finally the visualization of moisture content of the withering leaves is realized.