Introduction
The idea of the precision fish farming concept (Føre et al., 2018) is the automatization of the farming process, which can help control, monitor, and document the biological processes of fish cultivation. One of the optimization ways of the precision fish farming concept is finding an opportunity to get the individualized information about fish. The farmers could have personalized information about each fish in the tank (cage). Identification of individual fish could open a lot of possible solutions for the fish farmers. With the individualized information about fish, the processes of fish production could be more profitable. It helps deceases early detection, which could predict the high mortality. The identification of individuals could be a substitute for the fish tagging method (Andrews, 2004). This method is harmful and stressful for the fish. Fish has to be caught for identification which is time-consuming (Rácz et al., 2021). Non-invasive fish identification (photo-identification) is cheap and faster. The critical part of for individual fish identification method development is the large fish individuals dataset. We collected a large dataset of four fish species and provided it as an open access database for the researchers all over the world to play with and test their own methods on our data sets.. Those data sets were used for developmentof our own approaches for the identification of fish individuals. With the use of those data, three papers were published. One paper (European seabass and common carp) is under submitted status.
Materials and methods
Database is available on our data management platform – bioWES (http://www.biowes.org/). bioWES is a platform for experimental data and metadata management. General protocol (individual fish identification) obtains four different protocols for four fish species. Dataset of four fish species were collected,: Atlantic salmon Salmo salar, Sumatra barb Puntigrus tetrazona, European seabass, and common carp (Fig.1). Each protocol has two subprotocols. One is a protocol with the original data (data collection) and another one is protocol with the processed data. Some protocols end with the final protocol which have a final paper which was published according to the used data. Below are chapters with the description of four experiments (protocols).
Atlantic salmon - A total of 328 fish were used for short-term identification to test the identification power of the pattern. Thirty of the fish were tagged with PIT tags and used for long-term identification to test pattern stability. A total of 4 data collections were performed over six months at two months intervals. Three types of data were taken in each session: lateral view images of the fish out of the water (in a photographic tent) and underwater (in a small aquarium) and iris of the fisheye. The NIKON D90 digital camera was used to take approximately eight images of each fish in RAW format. Images of the 30-tagged fish were also collected as the long-term dataset during the next six months. The regions of interest (ROIs) were automatically extracted for all images in all datasets using image processing methods.
Sumatra barb - 25 fish individuals of Sumatra barb were used in this experiment. The digital camera with controlled lighting, the background, and the fish position, was used for data collection. Data were collected under different angles and positions; images were taken from one side view of all fish. Data were collected two times during two weeks for fish inside the aquarium. Tree images of each individual were taken in every data collection. The fish detection procedure consists of standard image segmentation based on color subtraction, object detection, filtration, and parametrization.
European seabass - Totally 300 sea bass were used during the experiment. All 300 fish were used for the testing the uniqueness of the visible patterns of the fish for identification of the individuals (short term experiment (ST)). Randomly selected 32 fish individuals out of the 300 fish were tagged by PIT tags for the long-term experiment (LT) to test the stability of chosen patterns during the cultivation period. After two months, the second data collection was performed. Four different ROI (regions of interest) were used for identification. All ROIs were detected automatically in the fish image.
Common carp - Four rounds of data collection were performed over the four-month experiment for carp identification. Thirty-two fish individuals were used in this experiment. Images were taken of the whole fish out of water. Each fish was caught in the cultivation tank, anesthetized in a bucket using clove oil, and moved to the green background for imaging. Automatic data processing consisted of three steps: fish detection, region of interest (ROI) selection, and feature extraction to describe the skin patterns of the ROI.
Results and discussion
The results of those experiments are fully presented in our papers. The datasets are scaled and focused on the different aspects of fish identification. The first dataset contains images of Sumatra barb. The methods of identification can be first tested on this dataset because of obvious stripe pattern on the fish and data close to the real conditions. We reached the the classification accuracy of 100%. More information is presented in the paper of Bekkozhayeva et al. (Bekkozhayeva, Saberioon, & Cisar, 2021). As a second datasetis the salmon data. Salmon also was strong dot patterns on the body but this dataset contaoins significantly more individuals. This identification resulst results had been described in the paper of Cisar et al. (Cisar, Bekkozhayeva, Movchan, Saberioon, & Schraml, 2021) and Schraml et al. (Schraml et al., 2020). Once the methods are successful dot strongly visinle pattern it can be tested on dataset for carp and sea bass. These fish species has only the scale pattern or lateral line patter which can be used for identification. But the results of European seabass and common carp are not published yet. The paper with the experiments and the results is under submission. In general, those studies demonstrated that the image-based individual fish identification is accurate and could be used for individual fish identification as a substitute for the tagging methods. The pattern stability was proved together with their stability.
Acknowledgments
The study was financially supported by the Ministry of Education, Youth and Sports of the Czech Republic - project „CENAKVA“(LM2018099), the CENAKVA Centre Development [No.CZ.1.05/2.1.00/19.0380] and GAJU 013/2019/Z.
References
Andrews, K. M. (2004). PitTaggingSimpleTechnology. 54(5), 447–454.
Bekkozhayeva, D., Saberioon, M., & Cisar, P. (2021). Automatic individual non-invasive photo-identification of fish ( Sumatra barb Puntigrus tetrazona ) using visible patterns on a body Content courtesy of Springer Nature , terms of use apply . Rights reserved . Content courtesy of Springer Nature , terms o.
Cisar, P., Bekkozhayeva, D., Movchan, O., Saberioon, M., & Schraml, R. (2021). Computer vision based individual fish identification using skin dot pattern. Scientific Reports, 11(1), 1–12. https://doi.org/10.1038/s41598-021-96476-4
Føre, M., Frank, K., Norton, T., Svendsen, E., Alfredsen, J. A., Dempster, T., … Berckmans, D. (2018). Precision fish farming: A new framework to improve production in aquaculture. Biosystems Engineering, 173, 176–193. https://doi.org/10.1016/j.biosystemseng.2017.10.014
Rácz, A., Allan, B., Dwyer, T., Thambithurai, D., Crespel, A., & Killen, S. S. (2021). Identification of individual zebrafish (Danio rerio): A refined protocol for vie tagging whilst considering animal welfare and the principles of the 3rs. Animals, 11(3), 1–18. https://doi.org/10.3390/ani11030616
Schraml, R., Hofbauer, H., Jalilian, E., Bekkozhayeva, D., Mohammadmehdi, S., Cisar, P., & Uhl, A. (2020). Towards fish individuality-based aquaculture. IEEE Transactions on Industrial Informatics, 1–1. https://doi.org/10.1109/tii.2020.3006933