Wild Cetacean Identification using Image Metadata

DEBORA POLLICELLI; CLAUDIO DELRIEUX; MARIANO COSCARELLA

JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY

SCIENCE PRESS

Lugar: Marrickville NSW; Año: 2017

1000-9000

Identification of individuals inmarine  species,  especially in Cetacea, isa critical  task in several biological and  ecological endeavours. Most of the times  this  is performed through human-assisted matchingwithin  a set of pictures taken  in different campaigns during several years and  spread  around wide geographical regions. This requires that the  sci- entists perform laborious tasks  in searching through archives of images, demanding a significant cognitive burden which  may be prone to intra-and interobserver operationalerrors.  On the other  hand,  additional avail- able  information, in particular the  metadata associatedto every  image,is not  fully taken  advantage of. The  present work presents the result ofapplying machine learning techniques over  the metadataof archives ofimages as an aid in the process of manual identification. The method was tested ona database containing several  pictures of230 different Com- merson?s  dolphins (Cephalorhynchus commersoni) taken  over  a span ofseven years.  A supervisedclassifier trained with identifications made  by the researchers was able  to identify  correctly above  90%of the  individ- uals  on the test set using only  the  metadata present in the  image  files.This  reduces  significantly the  number of images  to  be  manually  com- pared,   and  therefore the  time  and  errors associated with  the  assisted identification process.