Tracking fish traffic and enabling individual face recognition in Atlantic salmon
The Atlantic salmon aquaculture industry in Norway is under pressure to improve the welfare and health monitoring of the animals in farms. Advanced camera technology and facial recognition-algorithms can potentially document each individual fish and monitor developments in their growth, health- and welfare. The facial recognition-algorithms needs operationally to be developed and tested in full scale rearing systems and under commercial farming conditions.
The objectives of the following experiment are to i) track and monitor fish traffic in two adapted snorkel cage(s) in full scale commercial conditions as they are wandering through sensor system for monitoring fish health and welfare and ii) generate quantified training data for machine learning facial algorithms.
To monitor the traffic of individual fish through the sensor/snorkel, and to also allow for their identification via underwater camera technologies, a sub-sample of 6 000 Atlantic salmon (Salmo salar L.) will be tagged with unique multicolour T-bar anchor tags (5 000 in one cage and 1 000 in another) out of a total current population of 308 981 fish. Tags are an adapted version of an existing, well established T-bar anchor tag that have been used to tag 100+ thousands of fish in a huge number of e.g., aquacultural and ecological studies.
The VKM (2016) risk assessment of tagging methods for farmed salmonids stated there can be a high probability of moderate welfare consequences for fish tagged with external marker tags. Welfare risks can be both short- and long-term. In the short-term the tagging procedure involves handling and anaesthetising, and the fish can experience pain because of the tagging (VKM, 2016). Long-term problems can be additional hydrodynamic drag (tag, biofouling) which can lead to skin irritation or wounds (VKM, 2016). These factors may increase the risk of poor health, growth and mortality. Fish will be monitored closely using underwater cameras and in manual health and welfare screening to check for any potential adverse effects of tagging.
Non-invasive individual-based fish recognition using cameras and facial recognition algorithms removes the need for external fish tagging in the future. This would be incredibly beneficial for the aquaculture industry and would allow for the adoption of individual-based welfare and health management for farmed salmon. The 3Rs have been considered throughout the experimental planning phase and will be addressed during the study. As the experiment monitors salmon as they wander through an adapted snorkel cage with a sensor unit, this excludes any possibility for replacement of experimental animals. In terms of refinement, two smaller scale pilot studies have already been carried out (FOTS application 11182 and 14928) and this study is full scale testing and development. When considering the reduction aspect of the 3Rs, the current study will use commercial numbers and densities, to generate data that is directly transferable and applicable to commercial farming conditions (addressing a potential 4th R, Relevance). The number of fish that will be subjected to tagging is to generate enough training data for traffic monitoring and to quantify the efficacy of the facial recognition algorithms.
The objectives of the following experiment are to i) track and monitor fish traffic in two adapted snorkel cage(s) in full scale commercial conditions as they are wandering through sensor system for monitoring fish health and welfare and ii) generate quantified training data for machine learning facial algorithms.
To monitor the traffic of individual fish through the sensor/snorkel, and to also allow for their identification via underwater camera technologies, a sub-sample of 6 000 Atlantic salmon (Salmo salar L.) will be tagged with unique multicolour T-bar anchor tags (5 000 in one cage and 1 000 in another) out of a total current population of 308 981 fish. Tags are an adapted version of an existing, well established T-bar anchor tag that have been used to tag 100+ thousands of fish in a huge number of e.g., aquacultural and ecological studies.
The VKM (2016) risk assessment of tagging methods for farmed salmonids stated there can be a high probability of moderate welfare consequences for fish tagged with external marker tags. Welfare risks can be both short- and long-term. In the short-term the tagging procedure involves handling and anaesthetising, and the fish can experience pain because of the tagging (VKM, 2016). Long-term problems can be additional hydrodynamic drag (tag, biofouling) which can lead to skin irritation or wounds (VKM, 2016). These factors may increase the risk of poor health, growth and mortality. Fish will be monitored closely using underwater cameras and in manual health and welfare screening to check for any potential adverse effects of tagging.
Non-invasive individual-based fish recognition using cameras and facial recognition algorithms removes the need for external fish tagging in the future. This would be incredibly beneficial for the aquaculture industry and would allow for the adoption of individual-based welfare and health management for farmed salmon. The 3Rs have been considered throughout the experimental planning phase and will be addressed during the study. As the experiment monitors salmon as they wander through an adapted snorkel cage with a sensor unit, this excludes any possibility for replacement of experimental animals. In terms of refinement, two smaller scale pilot studies have already been carried out (FOTS application 11182 and 14928) and this study is full scale testing and development. When considering the reduction aspect of the 3Rs, the current study will use commercial numbers and densities, to generate data that is directly transferable and applicable to commercial farming conditions (addressing a potential 4th R, Relevance). The number of fish that will be subjected to tagging is to generate enough training data for traffic monitoring and to quantify the efficacy of the facial recognition algorithms.