Technology

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

ParameterSpecification
Image formatJPEG, TIFF, PNG
ResolutionMinimum 1 cm/pixel recommended
GeoreferencingEXIF GPS or sidecar files
Flight altitude30-80m 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