Improving the Quality of Green Tea Leaves in Sri Lanka Using AI

Sri Lanka is one of the leading Tea producers in the world. The tea industry contributes to more than 5% of the national GDP. The tea manufacturing process is a pipeline process and the total output of the tea production relies entirely on the quality of green leaves received at the tea factory. Unfortunately, a significant amount (10% - 40%) of total leaves get damaged during the collection, packing and transportation process.

The value chain starts with the tea leaves being plucked by humans to be stored in a basket or a special sack. Once tea leaves are plucked, the leaves are collected and packed into separate linen sacks and delivered to a factory by lorries. At the factory, the leaves get weighed by a supervisor and scattered to the special turf to start the weathering and fermentation processes.

The initial quality of tea leaves is checked at the factory through which the tea leaves get classified into 3 main categories:

1. Best

2. Below Best

3. Poor

The quality checking process is conducted by a technical manager or a tea inspector appointed by the Sri Lanka Tea Board. Good quality tea must have a minimum of 60% of ‘Best’ tea leaf classifications.

The quality check is a lengthy and time consuming process that needs to be completed prior to every tea manufacturing cycle. The price will be decided based on the above quality parameters. time-consuming

Identifying the type of tea leaves at the initial step of the production process using cutting-edge technologies to improve efficiency. 

Improve the quality of the tea using state of art technologies and prepare a transparent communication platform to share the information among factory owners, tea inspectors, and green leaf collectors. stakeholders in the tea value chain system. 

This tea manufacturing process can be exponentially expedited with the help of an Deep Learning based image classification solution. This solution will help to increase the efficiency ,transparency and profitability of the tea value chain process.

• Collect data and conduct feature engineering methods.
• Identify suitable machine learning / deep learning models for tea leaves classification.
• Train neural networks (model training).
• Validate models for training and testing for better results.
• Develop a portable device solution.
• Integrate the mobile device solution with the trained models.
• Apply the solution to the real environment and fine-tune the application.