Using computer vision to assess changes in ventilation rates of farmed Atlantic salmon ( Salmo salar ) following a freshwater bath

Computer vision is becoming a valuable tool for non‐invasive, continuous monitoring of farmed fish health and welfare, reducing the reliance on manual inspections. Field validation is needed to interpret computer vision outputs and assess their use in aquaculture. Freshwater baths, routinely applied to manage amoebic gill disease (AGD) in farmed Atlantic salmon ( Salmo salar ), improve gill health by reducing gill mucus. In this study, we aimed to determine whether a computer vision method could detect treatment‐related changes in ventilation after freshwater treatment. The method detected fish heads and classified mouth states (open or closed) using a convolutional neural network, followed by a tracking‐by‐detection approach to calculate mouth movement durations and estimate ventilation rates. Ventilation rates were calculated from 47 pairs of videos filmed before and after fish were bathed in freshwater at commercial farms in Tasmania, Australia. Additional data on water temperature and dissolved oxygen levels were recorded at the pens. A multiple linear mixed‐effects model was used to analyze associations between these variables. On average, median ventilation rates decreased by 2.56 open‐closed mouth cycles per minute after the baths. Our results contribute to understanding how computer vision could support the assessment of fish responses to freshwater treatment, which is commonly used to manage gill health issues such as AGD.