Application of Artificial Intelligence to the photo-identification of common bottlenose and Risso’s dolphins

Photo-identification of cetaceans is labor-intensive and time-consuming particularly in the case of large studies when manually performed. To that regard, Artificial Intelligence (AI) can support photo-identification studies with an extensive variety of statistical methods. The unique element of this work is the development of an AI-based system which makes decisions in terms of photo- identification not differently than a human mind would, thus providing users with an automated dorsal fin cropping and an individual-recognition pipeline for cetaceans. Both machine and human intelligences process symbols to interpret and learn from data. The developed system automatically identifies two categories of symbols: a) internal descriptors on the dorsal fin surface and b) outline descriptors, which are key-points over the fin contour. Internal and outline descriptors are both used for individual recognition in the classification process. The species of interest are the common bottlenose dolphin Tursiops truncatus and the Risso’s dolphin Grampus griseus. Sighting data have been acquired in the Gulf of Taranto in the Northern Ionian Sea (North-eastern Central Mediterranean Sea) and in the coastal waters around Pico Island in the Azores Archipelago (Eastern Atlantic Ocean). The accuracy of the dolphin photo-identification, computed by the proposed system, varies between 85% and 95%. Experimental results highlight that the developed automated system supports the work in terms of photo-identification of dolphin species, as essential prerequisite for insight studies on their spatial distributions, habitat uses, residency and migration patterns. Moreover, to make the proposed system accessible to a wider users’ community, we have also invested on the development of a well- documented graphical interface, together with a digital platform to smartly collect data