Technology · Artificial intelligence
Models that fly with the drone
KEA runs computer-vision and analytics models on board, in flight. That keeps latency low, keeps raw imagery on the aircraft and turns every mission into structured information rather than a folder of photos.
Three families of models
Navigation, analysis and situational awareness use different models, trained and validated separately.
Navigation models
Row detection, obstacle segmentation and terrain understanding for the autonomy stack.
Available
Agronomic models
Vegetation indices, stress and disease detection, canopy and cluster metrics.
In development
Situational models
Thermal anomaly detection, person and vehicle detection for security and emergency use.
In development
Edge inference on NVIDIA Jetson
Models are optimised for the on-board compute module and run in real time alongside the navigation software. Results – not raw video – are streamed as telemetry to the operator and the fleet platform.
- ComputeNVIDIA Jetson class module, GPU-accelerated inference
- InputsRGB, multispectral, thermal, LiDAR, RTK position, IMU
- OutputsGeo-referenced detections, indices, maps, alerts
- ValidationPer crop, region and season with pilot partners before release
Our AI principles
- Recommendations, not autonomous decisions: an agronomist or operator confirms every action.
- Transparent status: every model is labelled available, in development or research.
- Data stays yours: customer imagery is processed and stored in the EU and never used to train third-party systems.
- Measured, not claimed: detection performance is reported per validation set, with known limitations.
Let’s talk about your fields, your site or your mission.
Tell our engineering team what you need to see from the air. We will be honest about what works today, what we are testing and what is still research.