Precision Irrigation with Federated Learning
(2024)Privacy-Preserving Agriculture AI
Duration: 12 months
Synopsis
Honours thesis combining decentralised AI with edge computing for sustainable agriculture. Developed a federated learning framework on Raspberry Pi devices to predict irrigation needs while preserving farmer data privacy.
The Making Of
🎯 The Challenge
Agriculture accounts for 70% of global freshwater consumption, yet 40% is wasted. Farmers hesitate to share sensitive crop data with central servers, and rural areas lack reliable internet for cloud-based AI.
💡 The Solution
Built a custom Federated Learning framework from scratch. Used Raspberry Pi devices as edge nodes running multi-layer perceptron (MLP) networks, with FedAvg aggregation written from scratch. Only anonymous model updates transmitted via SSH, keeping raw data local. Tested on Autonomous Greenhouse Challenge data for tomato production.
🚀 The Impact
Achieved R² of 0.6-0.7 for irrigation prediction, close to centralised baseline of 0.7-0.8, while guaranteeing 100% data privacy. Co-supervised by Accenture through an industry placement scholarship, and awarded First Class Honours.
Key Metrics
Tech Stack (Cast & Crew)
- NumPy / PandasData & Modelling
- Raspberry PiEdge Devices
- Federated LearningPrivacy-Preserving ML
- PythonImplementation
- SSHSecure Communication
Quick Info
- Year
- 2024
- Duration
- 12 months
- Complexity
- Principal
- Featured
- Yes