Two-Stage AI Detection System
AI-BIRD employs a sophisticated two-stage approach to accurately detect and classify birds in high-resolution drone imagery.
Stage 1: Detection
Our detection model identifies individual birds within complex aerial imagery:
- High recall — >85% detection rate
- Resolution flexible — Works across flight altitudes
- Robust — Handles varying lighting and backgrounds
- Fast — Real-time processing capability
Stage 2: Classification
Once detected, each bird is classified to species level:
- >85% accuracy across target species
- Multi-species — Trained on gulls and terns
- Confidence scoring — Uncertainty quantification
- Continuous learning — Model updates with new data
Technical Specifications
Input Requirements
| Parameter | Specification |
|---|---|
| Image format | JPEG, TIFF, PNG |
| Resolution | Minimum 1 cm/pixel recommended |
| Georeferencing | EXIF GPS or sidecar files |
| Flight altitude | 30-80m AGL (adjustable) |
Output Formats
- GeoJSON — Point locations with species labels
- CSV — Tabular export for analysis
- PostGIS — Direct database integration
- API — RESTful endpoints for automation
Deployment Options
Cloud (AWS)
- Scalable processing
- API access
- Managed infrastructure
- Pay-per-use pricing
On-Premise
- Data sovereignty
- Custom integration
- Offline capability
- Enterprise licensing
Integration
AI-BIRD is designed for seamless integration with existing workflows:
- RESTful API — Submit images, retrieve results
- PostGIS — Native geospatial database support
- QGIS Plugin — Direct integration (coming soon)
- Python SDK — Programmatic access for automation