Computer Vision and Deep Learning Applied to the Photo-identification of Cetaceans

Photo-identification is the non-invasive process of uniquely identifying an individual among a set of individuals, based on the analysis of one or more photos. This is a specific task in cetaceans’ abundance and distribution studies, which can be effectively automated using computer vision and deep learning algorithms in large-scale studies. In this chapter, recent advances in the photo-identification of Risso’s dolphins are presented, covering the process from manual approaches to modern deep learning techniques. This manuscript highlights the strong multidisciplinary approach that is mandatory to accelerate and bring innovations working in multiple domains (marine biology and computer science in this case study). Particular attention is also given to the importance of data sharing, especially because it can be seen as a mandatory step that enables the proficient use of modern deep learning approaches to photo-identify a specimen. In the first part of the chapter, we present the state-of-the-art methods currently applied to the photo-identification task; the second part is devoted to describing the Smart Photo-Identification of Risso’s dolphins (SPIR) methods developed by our research team. Finally, future perspectives and directions of this research are discussed.