Geospatial Artificial Intelligence For Animals

Image

Overview

Monitoring whales is of broad interest to governments, academics, and industry around the world. Scientists employ a variety of research platforms (aerial, vessel, passive acoustic) to monitor the abundance and distribution of marine animals and update stock assessments. Different survey platforms have their own strengths and challenges. If we can also detect whales using very high resolution (VHR) satellite imagery we can address some of those challenges as well as add new data. 

Several recent publications have demonstrated the technological feasibility of identifying whales from VHR satellite imagery. NOAA is creating an operational system using artificial intelligence and cloud computing. This broad scientific collaboration is led by researchers from our Northeast Fisheries Science Center, Alaska Fisheries Science Center, and Alaska Regional Office, with colleagues at:

• U.S. Naval Research Laboratory
• Microsoft AI for Good Research Lab
• Bureau of Ocean Energy Management
• U.S. Geological Survey
• British Antarctic Survey
• University of Edinburgh
• Maxar Technologies

Read more about using artificial intelligence and satellites to monitor marine animals from space

Progress to Date

We formed our network of collaborators in 2020, and took a deep dive into the nuances of handling VHR satellite imagery. We tasked VHR imagery from WorldView-3, WorldView-2, and GeoEye satellites over seasonal aggregations of North Atlantic right whales and the Cook Inlet beluga whale. 

The British Antarctic Survey has developed detailed protocols for manually annotating the imagery to build up a standardized dataset. The Naval Research Laboratory and Microsoft AI for Good have been working on automated methods of image processing, and Microsoft AI for Good has developed a web-based annotation tool.

Read more about annotating VHR satellite imagery: a whale case study

Next Steps

We plan to harness these new annotation protocols and tools to develop an AI-ready dataset of whale detections in preparation for training machine learning models, and then to develop an operational system to process new imagery with these models.