| The processing of green tea mainly includes spreading,fixation,rolling,drying and other processing technology.The moisture content is an important basis for setting the process flow and process parameters and controlling the physical state change and chemical reaction process of raw materials.However,at present,most tea companies mainly rely on empirical methods and moisture analyzers to determine the moisture content in the green tea processing process,and cannot achieve rapid,non-destructive and accurate detection of moisture content,resulting in inconsistent tea quality.In order to realize the accurate measurement of moisture content in the green tea processing,it is necessary to rely on the oven drying method,but this method takes a long time,obviously cannot meet the needs of rapid perception,and is difficult to apply in actual production.Therefore,a rapid and accurate nondestructive testing method is urgently needed to detect the moisture content in the process of green tea processing.In this thesis,tea samples of different varieties were selected as the experimental objects,and the near-infrared spectroscopy technology,combined with machine vision technology and chemometric methods,was used to establish the moisture content prediction and discrimination model of green tea under different processing techniques,in order to improve the quality of finished green tea,Promote the standardization and intelligent development of green tea processing to provide a theoretical basis.The main research contents of this thesis are as follows:(1)Research on the moisture content of green tea in the green tea spreading process based on nearinfrared spectroscopyMoisture content is an important indicator in the process of green tea spreading.In order to realize the rapid and effective detection of the moisture content during the spreading process,this thesis collected 200near-infrared spectral data during the spreading process.In order to eliminate the noise information in the spectrum,five different preprocessing methods were adopted,and the PLSR prediction model was established to optimize the best preprocessing method.In order to reduce the redundant information in the spectrum and improve the prediction accuracy and speed of the model,the characteristic wavelengths in the spectrum were extracted based on the best preprocessing method,and the partial least squares regression(PLSR)and support vector regression(SVR)prediction models were established,and the effects of the models established based on different characteristic wavelength screening methods were analyzed,and determined VCPA-GA+SVR as the final prediction model of moisture content in green tea spreading process.The Rc and Rp of the model are both higher than 95%,and the RPD is greater than 2,indicating that the model has good prediction performance and can accurately predict the moisture content in the process of green tea spreading.(2)Establishment of a prediction model for the moisture content of green tea fixation leaves based on multi-source information fusionIn order to realize the rapid detection of moisture content in the green tea fixation process,a quantitative prediction model of moisture content change in the process of green tea fixation was constructed by using machine vision and near infrared spectroscopy.By collecting the spectrum and image information in the fixation process,and then fused the data information from the two sensors,using competitive adaptive reweighted sampling(CARS),variable combination population analysis(VCPA),variable combination population analysis and the iterative retained information variable algorithm(VCPAIRIV)and random frog(RF)algorithm to extract the characteristic wavelengths in the spectrum and combined the 15 colors and texture features in the image,the prediction models of partial least squares regression(PLSR)and support vector regression(SVR)were established.The results show that the model based on fusion data can effectively improve the prediction accuracy.Among them,the spectral feature wavelength extracted based on the CARS algorithm was combined with 15 color features of the image,the established SVR model had the best effect,in which the correlation coefficient of the calibration set(Rc)was 0.9742,prediction set correlation coefficient(Rp)value was 0.9719,and relative percent deviation(RPD)value was 4.1546,indicating the model had excellent predictive performance.In conclusion,this study proves the feasibility of integrating spectrum and image technology to predict the moisture content in the process of green tea fixation,overcomed the problem of low prediction accuracy of a single sensor,and laid a theoretical foundation for realizing the rapid nondestructive detection of the moisture content of green tea fixation leaves and accurately controlling the fixation quality.(3)The establishment of a green tea drying degree discrimination model based on near-infrared spectroscopy.Taking the green tea products in different drying stages as the research object,a discriminant model of green tea drying degree was established.According to the moisture content of dried samples,they were divided into three categories(under-drying,moderate drying,over-drying).By extracting the spectral absorbance values of tea samples with different drying degrees,combined with four spectral pretreatment methods,a partial least squares discrimination analysis(PLS-DA)model was established,and it was concluded that the discrimination model based on mean center(MC)pretreatment method has the best effect.Then,the competitive adaptive reweighted sampling(CARS),variables combination population analysis(VCPA),variables combination population analysis combined with iterative retained information variable algorithm(VCPA-IRIV)and variables combination population analysis combined with genetic algorithm(VCPA-GA)were used to screen the optimal characteristic wavelength related to moisture content.Then,combined with principal component analysis(PCA),k-nearest neighbor(KNN)and extreme learning machines(ELM)discrimination models were established.By analyzing the discrimination effects of different models,it is determined that MC+VCPA-IRIV+PCA+ELM has the best discrimination effect,and the discrimination effects of correction set and prediction set both reached 100%,which can realize the accurate discrimination of green tea drying degree.It lays a theoretical foundation for the development of online discrimination equipment for the drying degree of green tea.(4)Research on online detection method of moisture content of finished green tea based on multisource information fusionIn the process of green tea processing,in order to make the prediction model more accurate and comprehensive to predict the moisture information of green tea in different processing stages,taking zhuyeqi tea as the research object,collect the near-infrared spectral data and image information in different processing stages(spreading,fixation,first-drying,carding,and secondary-drying).By fusing the data information of the two sensors,the CARS was used to extract the characteristic wavelength in the spectrum,and combined with 9 color features and 6 texture features of the image,a nonlinear SVR prediction model was established.The results show that the SVR model based on data fusion has the good effect,the Rc value was 0.9804,the Rp value was 0.977,and the relative percent deviation(RPD)was 4.5002.It is shown that the data fusion based on machine vision and near-infrared spectroscopy technology can effectively predict the moisture content of green tea during the whole processing process from fresh leaves to dry finished product,and compared with the results of the moisture content prediction model for each processing stage of Gaoqiao Yinfeng tea. |
PDF Full Text Request