discussion / AI for Conservation  / 31 December 2015

Deep Learning Image Recognition of Species In Global Wildlife Crime Reporting

Over the past year the open data GDELT Project (http://gdeltproject.org/), which monitors local news coverage worldwide in 100 languages (65 languages are live machine translated in realtime), has been exploring global reporting on wildlife crime, using massive machine translation to catalog global poaching, illegal fishing, wildlife trafficking, and related topics, such as this map for Foreign Policy magazine mapping 30,000 articles in 65 languages over the first three months of this year:

http://foreignpolicy.com/2015/06/12/can-you-use-big-data-to-track-an-elephant-poacher/

Most recently we've begun exploring the application of deep learning image recognition to understanding the visual narrative around wildlife crime and cataloging the species found in news imagery, using content-based georeferencing to estimate the location the photo was taken, transcribe recognizable labels such as signs, flag weaponry and violence in the image, and many other annotation tasks using the Google Cloud Vision API.

Some initial demonstration applications for poaching, disaster response, pollution/littering, and georeferencing:

http://blog.gdeltproject.org/deep-learning-image-identification-to-counter-poaching/

http://blog.gdeltproject.org/deep-learning-triaging-for-disaster-response/

http://blog.gdeltproject.org/image-based-georeferencing-recognizing-locations-from-images/

http://blog.gdeltproject.org/48-hours-of-pollution-and-littering-around-the-world-through-deep-learning/

http://blog.gdeltproject.org/triaging-disaster-imagery-cyclone-pam-vanuatu/

The ALPHA live annotation stream is now available:

http://blog.gdeltproject.org/announcing-the-new-gdelt-visual-global-knowledge-graph-vgkg/

For more information, see the GDELT Project website (http://www.gdeltproject.org/) or the GDELT Blog (http://blog.gdeltproject.org/).