AZira — On-Device Pest Recognition
- Released
Overview
A Flutter application that identifies crop pests from a photograph in seconds and returns treatment guidance in Azerbaijani, English and Russian. The model runs on the phone itself, so the app stays fully functional in fields with no signal.
Project context
Built for Azerbaijani agriculture; the offline-first requirement comes from the fields themselves, where coverage is unreliable.
Key features
- On-device inference with a FastViT T12 model through TFLite — no server call, no signal needed
- Library of 20+ agricultural pests with descriptions in AZ, EN and RU
- Treatment guidance per pest: chemical options and agrotechnical advice
- PDF report generation and sharing for a detected pest
- Local scan history kept on the device
Technical details
- GitHub Actions — APK and AAB built and published on every v* tag
- Shared Preferences (offline cache)
- Dart
- TFLite (FastViT T12)
- Flutter 3.35.2
- Riverpod