Automated computer vision based individual salmon (Salmo salar) breathing rate estimation (SaBRE) for improved state observability

Salmon farming plays an important role in the Norwegian food industry, supplying a large portion of the world's salmon. Despite its economic importance, salmon farming faces unique biological challenges impacting the health and welfare of the fish. These factors limit the growth of salmon and the full economic potential of salmon farms, highlighting the need for advanced techniques in aquaculture to enhance productivity and sustainability. A key technical challenge is the unavailability of effective monitoring tools, leading to greater difficulties assessing the condition of salmon in water compared to the assessment of the animal stock on land-based farms. To improve the observability of salmon farms, we designed and created a computer-vision based approach for salmon breathing rate estimation (SaBRE), which allows the automatic monitoring of the respiration frequency of each individual salmon in a group of fish, seamlessly covering the entire workflow from video-stream input to the final data output (end-to-end). We thoroughly evaluated the capabilities of our method in two ways. Firstly, we performed a quantitative analysis of the constituent modules of SaBRE, revealing that all modules were highly accurate, including a salmon re-identification module that achieved an accuracy of 99.51 %. Secondly, we analyzed data from a salmon experiment with SaBRE, demonstrating that our algorithm provides high-quality respiration frequency information that can be compared with other types of experimental data to infer biological relationships. For the fish in our experiment, we observed that the ranking of the respiration frequencies in individual salmon remains relatively unchanged both in the short term and over longer periods (i.e., a salmon with a high breathing rate consistently remains among those with the highest rates, regardless of changes in the environment). Furthermore, a significant negative correlation (Pearson correlation coefficients with r values between −0.61 and −0.90 and p values below 0.01) was observed between our algorithm's estimated respiration frequency and the dissolved oxygen (dO 2 ) content in the water. In addition, the average breathing rate of the salmon was observed to increase in response to incidents potentially causing stress. • The individual salmon breathing rate in a group of 7 can be automatically monitored. • Breathing rate increases measurably with a decrease of dO2. • Breathing rate increases measurably when objects are inserted into a tank. • The relative breathing rate between fish in a group is consistent.