Basin-Scale Marine Mammal Monitoring: A Case Study Using Artificial Intelligence and Cloud Computing to Track Humpback Whales from their Breeding to their Feeding Grounds
PI: Berchok, Catherine (NOAA Alaska Fisheries Science Center)
Co-PI(s): Allen, Ann (NOAA Pacific Islands Fisheries Science Center) : Friday, Nancy (NOAA Alaska Fisheries Science Center) : Oleson, Erin (NOAA Pacific Islands Fisheries Science Center) : Woodrich, Dan (University of Washington)
Start Year: 2020 | Duration: 2 years
Partners: NOAA, Office of Naval Research, Google, University of Washington
The NOAA Pacific Islands Fisheries Science Center partnered with Google Artificial Intelligence to develop a deep machine learning model that recognizes humpback whale song in their decade-long passive acoustic dataset, a task that would have required hundreds of manual human hours. Here, we propose to apply this Google model to detect humpback song in recordings from the northern feeding grounds of this species, collected over the same decade by the NOAA Alaska Fisheries Science Center. This application will serve to test the transferability of the model to a novel data set, with retraining of the model proposed if necessary. We propose to further leverage the tools created by Google to develop artificial intelligence and machine learning methods for finer-scale comparison of humpback song structure, an indicator of population connectivity, in order to investigate broad scale humpback occurrence patterns across the entire North Pacific, and inform study of humpback populations worldwide. This project will establish a collaboration between NOAA science centers that will broaden usage of artificial intelligence for passive acoustic monitoring, as well as further an important NOAA collaboration with industry; both are essential steps in improving the utilization of cutting edge artificial intelligence and machine learning tools at NOAA.