Download Resume
Precision Irrigation with Federated Learning

Precision Irrigation with Federated Learning

(2025)

Privacy-Preserving Agriculture AI

AI/MLIoTResearch
☕☕☕☕☕
Coffee Required
Principal
Complexity
⭐⭐⭐⭐⭐
Documentation
👍👍
Would Build Again

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 MLP networks in PyTorch. 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. Published findings with industry partner Accenture.

Key Metrics

0.6-0.7
R² Score (Federated)
0.7-0.8
R² Score (Centralised)
100%
Data Privacy

Tech Stack (Cast & Crew)

  • PyTorchML Framework
  • Raspberry PiEdge Devices
  • Federated LearningPrivacy-Preserving ML
  • PythonImplementation
  • SSHSecure Communication

Quick Info

Year
2025
Duration
12 months
Complexity
Principal
Featured
Yes
← Back to All Projects