AI Algae and Jellyfish detection

fully autonomous operation

Partnering with HyperSpectral, Vestflow proposes a turnkey solution featuring an autonomous rugged drone platform equipped with a multi modal sensor suite to continuously scan the vicinity of each salmon farm. The system will integrate with HyperSpectral's SpecAI™ cloud analytics platform for real time detection and alerts, delivering critical metadata and insights to enable location managers to act quickly on early warnings. By "illuminating invisible threats and unlocking valuable insights" through AI powered spectroscopy, this approach offers faster, more cost effective detection than traditional manual sampling methods.

Algae Blooms

  o Detected via hyperspectral signature analysis - elevated chlorophyll-a will reflect more in green wavelengths and absorb more in blue/red, yielding 

     distinctive spectral curves
  o AI models trained on NOAA's extensive algae dataset will classify blooms (e.g., differentiating harmless phytoplankton from harmful Karenia (red

     tide) or cyanobacteria)
  o System measures bloom surface coverage area, concentration/severity via spectral indices, and estimated depth
  o Example metadata output: "HAB detected ~500m NW of Cage 4, covering ~30,000 m²"

Jellyfish

  o Identified through RGB/thermal detection
  o Many jellyfish appear as translucent disks visible from above in clear conditions
  o Studies confirm drones are effective for jellyfish surveying, matching or exceeding the accuracy of boat-based counts
  o System quantifies any jellyfish bloom - outlining swarm area on map and estimating count via AI pattern recognition

Sea Lice

  o Monitored indirectly through environmental conditions that favor lice outbreaks
  o System detects algae blooms and jellyfish that stress fish, preventing conditions where lice thrive
  o Platform is extensible - can integrate data from underwater cameras or lice count systems into SpecAI™

Data Capture and Processing

  • Regular monitoring with multiple daily flights (2-3 per day, more during high-risk seasons), adaptable based on risk levels
  • Drone autonomously surveys a 2km radius around each farm, covering ~12 km² in each flight

  • New areas will be monitored for complementation of database if necessary
  • Onboard edge computing provides initial preprocessing via several strategies:

        o Region of Interest (ROI) Streaming: Edge logic identifies areas with spectral signatures of algal blooms and sends only those portions
        o Compression and Binning: Hyperspectral data compressed by binning spectral bands where high detail isn't needed
        o Onboard AI Filtering: Confidently "no threat" areas are filtered to conserve bandwidth

  • Processed data transmitted to cloud over limited bandwidth connections (cellular/satellite)

SpecAI™ Analytics Platform

  • Hardware-agnostic platform ingests multi-sensor data from diverse inputs
  • AI models trained on NOAA's 25TB HAB dataset (covering years of weekly airborne hyperspectral scans over Lake Erie's blooms)

  • New areas will be monitored for complementation of database if necessary
  • Real-time threat detection with rich metadata including:

        o GPS-tagged threat locations
        o Bloom surface area calculations
        o Estimated depth of threat
        o Water temperature and other environmental parameters

• Dashboard and alerts delivered to managers with actionable information in minutes

Sample alert format: