Abstract:
Sri Lanka is renowned worldwide for its tea production, with Low Country
black tea grade playing a pivotal role in driving the nation‟s economy.
Identifying black tea grades is essential for quality assurance, pricing, and
standardization, enhancing the market demand and preserving tea production
quality. Traditional black tea grading methods in tea factories rely on manual
labor and subjective evaluation, leading to inconsistencies and inefficiencies
that can impact quality and pricing. This study introduces a deep learning
and Image Classification approach to automate black tea grading, focusing
on four main black tea grades: OPA (Orange Pekoe A), BOP (Broken Orange
Pekoe), OP1 (Orange Pekoe One), and P (Pekoe). For this research, images
were collected from multiple tea factories in different low-country regions,
focusing on factors identifying features such as dry tea leaves color, shape,
and size. To ensure accuracy, an 80/20 training and testing split was applied
to a dataset of approximately 1,000 images, maintaining a 1:1 ratio, allowing
for the clear identification of the components of a single dry tea leaf. After
preprocessing the dataset to an image resize of 80x80 pixels and labeling,
four CNN models that are Alex Net, VGG16, VGG19, and a custom CNN
model were trained and tested. Each model performed well, among which
the VGG16 model achieved the highest accuracy of 99.69%, confirming a
high ranking among the four black tea grade samples, which is a great
approach to automating the identification process. This automated grading
framework reduces human error and labor while providing a scalable and
objective solution for the tea industry, leading to more consistent grading
standards. This CNN-based automated grading framework enhances quality
assurance and production efficiency while reducing human error and labor. It
provides a scalable and objective solution for the tea industry, paving the
way for consistent grading standards and supporting Sri Lanka's tea sector in
maintaining competitive, high-quality production.
