Automate Tea Clone Identification using Convolutional Neural Networks

Abstract

"One of the greatest and most widely consumed teas in the world is Sri Lankan tea. Tea Research Institute (TRI) in Sri Lanka has developed numerous varieties of tea through cloning, known as the TRI series clones. These clones have distinct physical features, although some share similarities, making manual identification challenging and prone to mistakes. An in-depth review of previous studies on tea clone classification highlighted existing shortcomings in this area. To address these issues, the study employed a Convolutional Neural Network (CNN) model, marking a significant improvement in identifying tea clones accurately. This research stands out as the first to categorize the newly introduced TRI 5000 series clones in Sri Lanka. It also introduces a unique dataset containing images of four tea clone leaves (TRI 5001, TRI 5002, TRI 5003, and TRI 5004). This opens opportunities for further research to include more TRI clone types not covered in this study. The effectiveness of this research was confirmed through rigorous testing with selected evaluation metrics. "