In the heart of Sri Lanka’s agricultural industry lies tea, a crop that not only defines the landscape but also plays a crucial role in the country’s economy. As a primary export product, tea contributes significantly to Sri Lanka’s GDP. However, this vital industry faces persistent challenges from tea leaf diseases such as Blister Blight, Gray Blight, and Black Blight, which threaten yield quality and production. Traditional methods of disease detection, reliant on manual observation, are time-consuming, labor-intensive, and often inaccurate. This calls for a smarter, more efficient approach—enter AI-driven solutions.

The Challenge: Disease Detection in Tea Farming

The manual process of identifying tea leaf diseases involves expert evaluation, which can be both subjective and inconsistent. In large-scale plantations, detecting diseases early enough to prevent significant damage is nearly impossible without specialized tools. These challenges underscore the need for a robust system that can automate disease detection, enabling early intervention and improved management of tea plantations.

Our Solution: AI-Powered Tea Leaf Disease Detection

Recognizing these challenges, we developed an innovative system that combines machine learning (ML) and computer vision to detect tea leaf diseases accurately and efficiently. This system utilizes two state-of-the-art ML models tailored for different operational needs:

MobileNetV1 for Offline Use

MobileNetV1, a lightweight Convolutional Neural Network (CNN), was selected for its balance of high accuracy (94%) and low computational requirements. This model was integrated into a mobile application using TensorFlow Lite, enabling offline functionality. Tea farmers can simply take a photo of a leaf, and the app instantly diagnoses the leaf’s health without requiring internet access.

Vision Transformers for Cloud Processing

Vision Transformers, known for their scalability and precision, were hosted on Hugging Face Inference API. This cloud-based model is designed for handling larger datasets and providing more complex analysis, making it ideal for advanced scenarios and centralized processing.

Backend and Deployment

The backend, built using Flask, facilitates seamless communication between the mobile app and cloud infrastructure. Hosted on Google Cloud Platform (GCP), this setup ensures reliability and scalability for both small-scale users and larger tea estates. And MongoDB was used as the database.

Features of the System

1. Real-Time Disease Detection

The mobile application offers real-time results by processing images either offline (via TensorFlow Lite) or online (via Vision Transformers). This versatility ensures accessibility in areas with limited internet connectivity.

2. User-Friendly Interface

Built using the Flutter framework, the app is designed with simplicity in mind. Even users with minimal technical knowledge can diagnose diseases by snapping a photo of the tea leaf.

3. GIS-Enhanced Insights

Geotagging features allow users to map the spread of diseases across plantations. By visualizing disease distribution profiles, plantation managers can target hotspots for intervention, optimizing the use of pesticides and resources.

4. Scalable Architecture

The system’s dual-model setup caters to diverse needs. MobileNetV1 is perfect for on-the-go use, while Vision Transformers handle more complex, cloud-based analyses.

Mobile Application

Impact on Tea Farming

This system revolutionizes tea farming by enabling detection of diseases, reducing the reliance on manual inspections, and empowering farmers with actionable insights. The offline functionality ensures accessibility even in remote plantations, while the GIS integration provides a strategic advantage in managing disease outbreaks.

With its accuracy, scalability, and ease of use, this solution promises to enhance the productivity and sustainability of Sri Lanka’s tea industry.

Looking Ahead

While MobileNetV1 and Vision Transformers are at the core of this project, future enhancements could include:

  • Scaling Vision Transformers: Expanding their application to larger datasets and additional crops.
  • Advanced GIS Features: Integrating predictive analytics for disease spread patterns.
  • Collaborations: Partnering with agricultural organizations to refine and deploy the system at scale.

Conclusion

This research represents a significant leap forward in agricultural technology, showcasing how AI can transform traditional farming practices. By combining cutting-edge machine learning with practical deployment strategies, we aim to secure the future of Sri Lanka’s tea industry while empowering farmers to embrace smart, sustainable practices.