Developing Computer Vision and Deep Learning Methods to Improve Aerial Surveys of Marine Wildlife

Submitted by atripp on

BOEM, U.S. Fish and Wildlife Service, and the U.S. Geological Survey are collaborating to foster research on deep learning methods that automate remote sensing data for wildlife population surveys. The Atlantic Marine Assessment Program for Protected Species (AMAPPS), in its third phase, is developing automated ways to rapidly filter and subset digital aerial imagery of marine birds, cetaceans, and sea turtles. A major challenge for integrating remote sensing methods for large-scale population surveys is the tremendous volume of data collected during image-based surveys and the lack of suitable tools for automated filtering of imagery, counting detected wildlife targets, and classifying wildlife targets. Computer vision, coupled with deep learning, shows great promise for automating detection and classification of wildlife from digital imagery. These agencies recognize a set of common wildlife population monitoring priorities and shared objectives to enhance employee safety and to improve data quality.

Examples of individual sea ducks detected and identified in a raft near Cape Cod, Massachusetts, USA. Individual birds were delineated with a rectangle and identified to the lowest taxonomic level. Additional attributes such as age, gender and activity were included when resolvable. Photo credit: Mark Koneff, USFWS


Detection of individual sea ducks in a raft near Cape Cod, Massachusetts, USA. Photo credit: Mark Koneff, USFWS


Author Name
Timothy White
Author Email