Classification of Sri Lankan leafy tea grades using deep learning

Abstract

Tea grading is a crucial step in the tea manufacturing process. The procedure is labor-intensive, subjective and time-consuming, relying heavily on the knowledge of experienced tea workers. To address these limitations, this research introduces an automated tea grade classification system using deep learning. The proposed method aims to classify five Sri Lankan leafy tea grades including BOP1, OP1, OPA, Pekoe and Chunmees based on images. A custom dataset containing 2,500 images was created for this purpose. Dynamic data augmentation techniques were applied to enhance the model generalization. Convolutional Neural Network (CNN), MobileNetV2 and VGG16 models were evaluated to identify the most accurate classifier. Among them, VGG16 achieved the highest accuracy of 99%.