Tea Quality Assessment using AI and Computer Vision with Transfer Learning and multitask learning

Abstract

The industry faces unnecessary challenges in quality assessment because the current measurement techniques and leaf aging evaluation methods demonstrate performance restrictions. Traditional grading methods depend on visual grading which demands long execution times but this method results in objective impacts to freshness determination and grade assignment. The developed automated tea quality evaluation tool serves to assist the tea industry in making decisions. The project solves this challenge by employing a transfer learning approach with additional training of pre-existing models against tea leaf picture collections for classification needs. The proposed design for the model implements automated multitask analysis for leaf grading alongside color and texture input-based ripeness evaluation. Quality grading and ripeness level determination acquire greater accuracy when researchers integrate histogram and moment features that include mean, variance, skewness and kurtosis. Prototype testing shows the model reaches high precision in classification through both exploration tests and measures supported by Confusion Matrix and AUC-ROC coefficients. This tested methodology should succeed in providing reliable tea quality assessment but researchers will continue to work on making the model more stable and adaptable for diverse tea category testing.