The Design Of A Flat Tea Shape Index Analysis Software And The Build Of Regression Models Of Dafo Longjing

Tea review by sensory evaluation requires high standard of professional skill, and also contants high subjective error. There are attempts to display the quality of tea by the content of polyphenols, theanine, aroma components, but these kinds of methods often need to be achieved in the laboratory, with higher operation skills, detection reagents, equipment, and so on. This study designed a software to analysis the flat tea shape Indicators, and built some recognition models with the Dafo Longj ing to achieve the recognition.(1) This software can directly read the indicators of tea length(L), width(W), the perimeter (P), the area(A), Red(R), Green (G), blue (B), and their means and standard deviation. It also can be associated with the database software, make the data warehousing screening, storage, read, and can be read out the image of each tea leaf and its index data for further screening.(2) Except the indicators directly read by the software, we choice some composite indicators to show more information of the tea. Like the Wj shows the overall width of a tea leaf, L/W, L/Wj to show the length and width ratio, Wj/W, P/Pj to show the bifurcation, and the brightness, saturation and hue.(3) Build Two types of discriminant model, three types of regression models with the standard sample of Dafo Longjing. The linear discriminant model and stepwise linear discriminant model show high classification rate, and the regression models also show the high correlation coefficients.(4) Test these models with another Dafo Longjing standard sample, three kinds of Dafo Longjing, standard sample of Xihu Longjing. The linear discriminant model and stepwise linear discriminant model show high classification rate, but only20%correct rate in the text with external samples. The multiple linear regression model built by the shape indicators which are significantly related to the sensory score of the tea shape shows the best correctness, with the average relative error of0.89%, from0.11%—1.76%. But the regression model with all the shape indicators gets the average relative error of13.86%, from2.06%—42.68%. The Stepwise linear regression model’s is9.97%, from0.87%—42.89%. So, the multiple linear regression model built by the shape indicators which are significantly related to the sensory score of the tea shape is the best.