Digital Blending Optimization Strategy For Tea Based On Multi-source Information Fusion

Tea blending,an essential process in tea processing and sales,plays a significant role in ensuring a stable tea quality,maximizing tea yield,and improving economic benefits.Professional blending masters traditionally rely on sensory evaluation results of various raw teas and past experience to develop a blend through repeated trials and improvements.However,this approach requires a high level of skill,is time-consuming,and the blending result is often subjective,making it challenging to keep up with the modern development trend of tea production.To address these challenges,this study used Yinghong No.9 black tea as the experimental material to construct a black tea sensory quality evaluation model,leveraging computer vision technology and near-infrared spectroscopy quantitative analysis technology.The model enabled accurate and fast digitized evaluation of the appearance and taste quality of black tea,and introduced multi-objective intelligent optimization algorithms to construct digital blending optimization strategies based on the quality and price of different raw teas.This allowed for the rapid formation of an optimal formula,digitizing the tea blending process.The study proposed an improved K-means clustering image segmentation algorithm applicable to tea image with flat-stacked acquisition,overcoming difficulties in effective image feature extraction,poor interpretability of features,and low processing efficiency for tea image segmentation.The algorithm achieved fast and accurate segmentation of the dark background area,tea area,and golden pekoe area in the black tea image,with an average processing time of only 6.935 s per image.The relationship between Key descriptions of golden pekoe,color and strip in the sensory evaluation of black tea appearance and digital image features were established through the L*a*b*color model and gray-level run-length matrix.Based on the extracted key image features,black tea appearance quality prediction models were built using random forest,support vector regression,and BP neural network,respectively.The results showed that the comprehensive performance of the random forest-based black tea appearance score prediction model was the best,with R_P~2,RMSEP,and RPD of the prediction set reaching 0.922,1.222,and 3.621,respectively.Furthermore,the study analyzed the distribution of tea polyphenols,free amino acid,phenol-amino acid ratio,catechins,soluble saccharides,caffeine and water extracts in black tea of different qualities,exploring the influence of the main flavor substances on the taste quality of black tea.Near-infrared spectroscopy quantitative analysis technology was used to model and predict the content of main flavor substances and taste scores in black tea.The prediction performance of the models under different spectral preprocessing methods and wavelength selection algorithms was analyzed and compared,showing that near-infrared spectroscopy could effectively reflect the chemical composition information of the main flavor substances in black tea.The feature wavelength extraction results under the optimal spectral processing algorithm combination could well preserve the related bands of black tea taste quality,and the prediction effect of various taste quality index prediction models was satisfactory.Finally,a digital blending optimization strategy based on image and spectral information for tea was proposed,transforming the tea blending problem into a multi-objective optimization problem that considers both cost and quality,and establishing a mathematical model for multi-objective optimization.The quality prediction results were applied to the digital blending of black tea,and a cost and quality deviation objective function for blending was constructed.The blending ratios of each raw tea were taken as decision variables,and various multi-objective intelligent optimization algorithms were used for automatic optimization of the formula,quickly forming a set of optimal blending solutions.The effectiveness of the generated blending schemes by the model was validated through blending experiments and sensory evaluation,demonstrating that the proposed digital blending optimization strategy can quickly form optimal blending solutions based on the quality and price of different raw teas,serving as an effective auxiliary tool in the blending process,providing guidance for the optimization of tea blending process.