AgroSaviour

In Ghana, crop diseases cost farmers a lot of losses yearly threatening food security and lives. Our solution puts real-time crop disease detection in the hands of every farmer using just a phone!.

Comment

🌱 Inspiration

In Ghana, agriculture is not just a sector rather it’s a lifeline. However, crop diseases and pests are silently crippling yields, livelihoods, and food security. Many farmers lack access to timely diagnosis or agricultural extension services. We were inspired to build a solution that democratizes plant health expertise using AI to empower even the most remote farmer with real-time, accurate disease detection and guidance.

🌾 What it does

AgroSaviour is a smart, AI-powered system that identifies 22 different crop diseases across four key Ghanaian crops such as cassava, cashew, maize, and tomato using a simple photo. Farmers can scan leaves using a mobile app or IoT camera module, and instantly receive disease name, confidence score, and actionable treatment recommendations. It also logs detection history and supports user-specific dashboards via Firebase integration.

🛠 How we built it

We trained a high-performance classification model using EfficientNet-B4 on a custom-built, 125K+ image dataset (CCMT), achieving over 98.6% accuracy. The backend was developed using FastAPI, supporting both .pt and .onnx model formats for flexible deployment. We integrated Firebase for authentication, user-based prediction history, and real-time database logging. The app interfaces with ESP32-CAM modules or mobile phones for live streaming detection and supports mobile and web clients.

🚧 Challenges we ran into

  • Converting a complex folder-based dataset into a unified format for training
  • Ensuring model generalization across diverse field conditions and crops
  • Building a modular, real-time pipeline that works seamlessly on both low-end devices and cloud deployments
  • Switching backend storage from Supabase to Firebase while preserving data models and logic

🏆 Accomplishments that we're proud of

  • Developed a robust 22-class crop disease model with industry-level accuracy
  • Built a scalable, production-ready API backend with Firebase user support
  • Integrated IoT (ESP32-CAM) for real-time detection
  • Created a user-centered, multilingual interface tailored for Ghanaian farmers

📚 What we learned

  • The importance of dataset quality and preprocessing in agricultural AI
  • Optimizing deep learning models for both accuracy and speed in real-world settings
  • The power of Firebase for secure, user-specific data tracking and app-scale deployments
  • Collaborative development, version control, and agile iteration from research to deployment

🚀 What's next for AgroSaviour

  • Adding symptom-based text diagnosis to support farmers without cameras
  • Expanding crop and disease coverage (e.g., cocoa, plantain)
  • Offline model support for remote farms with no internet
  • Collaborations with agricultural ministries, NGOs, and cooperatives
  • Launching an early-warning system using satellite and climate data

Built With

Share this project:

Updates