Data for non-invasive (photo) individual fish identification of multiple species

Dinara Bartunek; Petr Cisar

doi:10.1016/j.dib.2023.109221

Data for non-invasive (photo) individual fish identification of multiple species

Data Brief. 2023 May 9:48:109221. doi: 10.1016/j.dib.2023.109221. eCollection 2023 Jun.

Authors

Dinara Bartunek¹, Petr Cisar¹

Affiliation

¹ Laboratory of Signal and Image Processing, Institute of Complex Systems, Faculty of Fisheries and Protection of Waters, CENAKVA, University of South Bohemia in Ceske Budejovice, Zámek 136, Nové Hrady 373 33, Czechia.

Abstract

This paper describes data from five studies focused on the individual fish identification of the same species. The lateral images of five fish species are present in the dataset. The dataset's primary purpose is to provide a data to develop a non-invasive and remote method of individual fish identification using fish skin patterns, which can serve as a substitute for the common invasive fish tagging. The lateral images of the whole fish body on the homogenous background for Sumatra barb, Atlantic salmon, Sea bass, Common carp and Rainbow trout are available with automatically extracted parts of the fish with skin patterns. A different number of individuals (Sumatra barb - 43, Atlantic salmon - 330, Sea bass - 300, Common carp - 32, Rainbow trout - 1849) were photographed by the digital camera Nikon D60 under controlled conditions. The photographs of only one side of the fish with several (from 3 to 20) repetitions were taken. Common carp, Rainbow trout and Sea bass were photographed out of the water. Atlantic salmon was photographed underwater, out of the water, and the eye of the fish was photographed by the microscope camera. Sumatra barb was photographed under the water only. For all species, except Rainbow trout, the data collection was repeated after a different period (Sumatra barb - four months, Atlantic salmon - six months, Sea bass - one month, Common carp - four months) to collect the data for a study of skin patter changes (ageing). The development of the method for photo-based individual fish identification was performed on all datasets. The identification accuracy for all species for all periods was 100% using the nearest neighbour classification. Different methods for skin pattern parametrization were used. The dataset can be used to develop remote and non-invasive individual fish identification methods. The studies focused on the discrimination power of the skin pattern can benefit from it. The changes of skin patterns due to fish ageing can be explored from the dataset.

Keywords: Automation; Computer vision; Fish individual identification; Fish lateral image; Machine learning; Non-invasive identification; Tagging.