ARIANNA: A novel deep learning-based system for fin contours analysis in individual recognition of dolphins

Starting from a digital image that represents the dolphin's body, distinctive features are extracted and used to find the identity of the unknown dolphin in a set of known individuals. This process is called photo identification, used by experts to monitor dolphins, providing relevant data to preserve the environment and its biodiversity. In this work, we show how semantic segmentation can be used to automatically extract a dolphin's fin contour starting from a cropped photo of the fin, and how this contour can be used for individual identification. A novel contour-based system, called ARIANNA, for the automated cetacean photo identification was designed, developed and tested. The novelty of this system is the adoption of two original modules. The first one, which takes as input a new cropped fin image of unknown dolphin, is devoted to the extraction of a mask that depicts the outline of the unknown fin; the core of this module is a trained neural network, specialized in semantic segmentation of images. The second module is designed to compare the outline of the unknown fin with the outlines of all known dolphins, collected in a referring catalogue, returning a ranked list of the best matches where to search the dolphin identity. The experiments were conducted on images collected between 2013 and 2020 in the Northern Ionian Sea (Central-eastern Mediterranean Sea), which presented cropped fins of Risso's dolphin Grampus griseus, one of the least-known cetacean species on a global and Mediterranean scale. The results suggest that ARIANNA provides advances over the state-of-the-art methods, can efficiently assist researchers in the photo identification of dolphins and can be a starting point for further studies on the photo identification of different species based on semantic segmentation.