Deep learning-based computer vision for assessing the palatability of dietary black soldier fly (Hermetia illucens) larvae meal in rainbow trout

Understanding the palatability and behavioural responses of fish to novel feed ingredients, particularly the commonly known black soldier fly (Hermetia illucens) larvae meal (HI), is essential for optimizing sustainable aquafeed formulations. This study evaluated the feeding behaviour of rainbow trout (Oncorhynchus mykiss) fed HI-containing diets using a computer vision-based monitoring system. Four isoproteic, isolipidic, and isoenergetic diets were formulated, including HI at 0, 2.5, 5, and 10% at the expense of fishmeal and extruded into 4 mm pellet diameter. Eighty-four rainbow trout (158.9 ± 3.0 g) were randomly distributed into twelve 50 L tanks (7 fish/tank) and assigned to one of four dietary treatments in triplicate. Fish were fed the experimental diets for five days to allow adaptation to both the experimental setup and the diets. On the sixth day, video recording and feed intake monitoring were conducted. Video footage was analysed using computer vision techniques to extract quantitative behavioural metrics, including pellet detection and counting, fish swimming speed estimation, and spatial movement patterns, which were processed using object detection and tracking algorithms. Nutritional composition and volatile organic compounds of experimental diets were determined. Feed intake was inferred from the consumption rate, and pellets disappeared over time. The findings indicate that including HI into trout diets significantly affects pellet acceptability and swimming dynamics. The computer vision approach provided highresolution, non-invasive behavioural data, enabling precise differentiation in palatability across diets. This study highlights the utility of digital tools in aquaculture research and offers novel insights into the feeding responses of fish to sustainable feed ingredients such as insect meal.