As energy development increases along the Atlantic Coast and across the Outer Continental Shelf (OCS), BOEM needs robust species and site-specific information on the seasonal distribution and abundance of marine species – including seabirds, marine mammals and turtles – that could be affected by offshore energy activities.
When it comes to identifying and counting multiple species on marine wildlife aerial surveys, observer accuracy can vary widely. BOEM intends to improve the detection and classification of species recorded on aerial surveys through the Atlantic Marine Assessment Program for Protected Species (AMAPPS) III project.
The U.S. Fish and Wildlife Service (USFWS) are conducting aerial surveys in BOEM-identified target areas to capture high-resolution images of marine mammals, sea turtles and seabirds to build a database of annotated photos. The database will train deep learning computer vision algorithms to count and identify the species and – once trained – deep learning models will automatically detect and count species in new images. BOEM will compare the accuracy of species identification using high-resolution imagery with previous AMAPPS observer-based data to quantify the performance of both methods.