Project Ocean Oasis advances reef conservation with Ultralytics YOLO

As marine ecosystems face unprecedented threats from climate change, pollution, and overfishing, the demand for scalable, evidence-based ocean intelligence has never been greater. Project Ocean Oasis (PO2), an Australia-based not-for-profit, is meeting this challenge head-on by building autonomous monitoring systems designed to operate for years in harsh marine environments.

By integrating Ultralytics YOLO models into its edge-to-cloud pipeline and leveraging Ultralytics Platform, PO2 is laying the groundwork for a new era of continuous, AI-powered reef and biodiversity monitoring.

Link to this sectionBuilding the foundation of ocean intelligence#

Project Ocean Oasis develops autonomous, AI-enabled systems for long-term ocean intelligence. Built in collaboration with WASSOC (a subsea engineering firm with over 40 years of experience), AWS, and Ultralytics, PO2 delivers an end-to-end platform combining rugged hardware, edge and cloud-based AI, and scientifically validated datasets to support evidence-based protection of marine ecosystems.

Each PO2 unit is designed to run multiple AI models simultaneously, with enough internal headroom for future expansion. The systems are engineered to withstand reef surges, corrosion, and biofouling, operating maintenance-free for years at a time and scalable across diverse environments.

Link to this sectionThe challenge of long-term, autonomous reef monitoring#

Monitoring reefs and marine biodiversity at scale presents a unique set of challenges. Traditional methods rely on divers, boats, and manual data collection, which are expensive, intermittent, and often disruptive to the ecosystems they study. As reef health declines globally, the gap between the data scientists need and the data they can realistically collect continues to widen.

Even as computer vision and edge AI become more accessible, deploying these technologies underwater introduces new constraints. Devices must be solar and battery-powered, capable of running multiple AI models in parallel, and rugged enough to survive years of saltwater exposure without active maintenance. Power efficiency is critical: With only around 600 watts of daily power available, every component, from the camera to the inference chip, must be carefully chosen to maximize uptime while delivering accurate, real-time insights.

For PO2, the missing piece was a model framework flexible enough to run on a wide range of edge hardware, accurate enough for marine biodiversity tracking, and easy enough to iterate on as the platform evolved across multi-year deployments.

Link to this sectionFrom edge to cloud with Ultralytics YOLO#

PO2's monitoring platform combines underwater cameras, hydrophones, and other sensors mounted on autonomous buoys equipped with solar panels, batteries, and satellite connectivity.

For Audio, the PO2 HydroPulse (custom edge embedded hydrophone) runs at around 1.3 W on commodity ARM hardware, low enough for long-term solar and battery deployment.

Looking to run on the cameras, Ultralytics YOLO models are looking to be deployed on edge hardware within the system, performing real-time object detection and tracking on video streams to identify and track fish and other marine life directly underwater.

YOLO tracking links the same individual across frames, so it is only counted once, with its highest-confidence detection taken as that fish's record. A single record is essentially:

{
  "label": "Acanthurus triostegus",
  "confidence": 0.94,
  "length_cm": 18
}

with session-level MaxN counts and environmental context attached. This results in standardised, science-ready data rather than footage to review later.

Fig 1. Ultralytics YOLO26 running species-level detection on Hawaiian reef footage.

Each camera unit will detect and track at the edge, then ship compact structured events (kilobyte-scale JSON), not raw video.

When the system detects an object of interest, the inference results are sent via satellite to a cloud architecture, where they are aggregated, analyzed, and surfaced to scientists and conservation partners. This edge-first design dramatically reduces transmission demands, conserves power, and enables the platform to operate for years per deployment without manual intervention.

To future-proof the system, PO2 worked closely with Ultralytics to evaluate the latest generation of NPU-powered edge AI accelerators. This collaboration helped PO2 narrow down the right hardware combinations for its strict power budget while maintaining the flexibility to add new AI models, such as audio inference on hydrophone data, as the platform grows.

Fig 2. Outlining Ocean Oasis’ solution.

Link to this sectionAn early adopter of Ultralytics Platform#

In addition to using Ultralytics YOLO models in production, PO2 became one of the very first enterprise users of Ultralytics Platform, the new end-to-end environment for annotating, training, and deploying YOLO models in one place.

Using Ultralytics Platform, PO2's team can manage marine datasets, train both standard and Ultralytics Enterprise YOLO26 models, leverage Smart Annotation to dramatically speed up data labeling, and export trained models to virtually any format for deployment across their edge hardware. This unified workflow has helped PO2 iterate quickly as the project evolves, while keeping data, models, and experiments centralized as the team scales.

Link to this sectionWhy choose Ultralytics YOLO?#

For Project Ocean Oasis, the collaboration with Ultralytics goes beyond model performance, providing access to expertise and a unified platform that supports the team at every stage of development. Ultralytics YOLO models offer the flexibility to run on a wide range of edge devices, from ultra-low-power MCUs to higher-performance NPU accelerators, while delivering the accuracy needed for biodiversity monitoring underwater.

By working closely with Ultralytics, PO2 was able to navigate the complex landscape of edge AI accelerators and identify the right hardware partners for their multi-year deployment goals.

Link to this sectionTowards a healthier ocean#

Project Ocean Oasis has validated its first sensors end-to-end, from edge AI inference through to the cloud dashboard, and is now preparing for in-water deployment. The hardware build is partner-ready, the cloud architecture is operational, and scientific investigators are in place at deployment sites.

From 2026 onwards, the team will scale across sites, expand the on-device AI suite, and lay the groundwork for an interlinked global monitoring network.

PO2 is built on a multi-generational time horizon. Continuous marine monitoring is the foundational layer, the evidence base that makes everything downstream possible. The work is not to finish it, but to start it well.

Interested in building Vision AI solutions of your own? Visit our GitHub repository to explore Ultralytics YOLO models, learn how YOLO is supporting innovations across AI in conservation, and check out Ultralytics Platform to start building your own end-to-end vision AI workflows.