Background Feed efficiency is a key trait affecting the economic and environmental sustainability of aquaculture, and accurate measurement of individual feed-intake is essential for evaluating feed efficiency and individual variation. Existing methods for measuring individual feed-intake in fish commonly rely on individual isolation, manual video analysis, or X-ray tracing, which makes it difficult to simultaneously maintain group-rearing conditions and achieve continuous, automated monitoring. Continuous automatic measurement of individual feed-intake under group-rearing conditions has not yet been well established for spotted seabass (Lateolabrax maculatus).

Objective This study aims to adapt and optimize an intelligent individual feed-intake measurement system for the rapid and transient feeding behavior of L. maculatus, and to evaluate the effects of the feeding device on growth and stress status.

Methods 30 fish were used in a 14-day feeding trial, with 15 fish in the individual isolation group and 15 fish in the group-reared control group, to compare feed consumption under the two conditions. An individual feed-intake measurement system was then optimized by improving the feed-intake recognition and calculation module, the structure of the restricted-access feeding station, and the temporal matching logic between passive integrated transponder (PIT) signals and image-based feed-pellet counts. After optimization, 472 fish were randomly assigned to a device group (using feeding stations) or a control group (without feeding stations), (236 fish per group) for a 27-day preliminary application trial. Growth performance, mortality and feed conversion ratio were recorded, and plasma cortisol was measured in five fish per group at the end of the trial.

Results Feed consumption in the individual isolation group was generally lower than that in the group-reared control group during the 14-day trial, suggesting that isolation may affect normal feeding. After optimization of the recognition module, feeding-station opening and counting logic, the accuracy of feeding-event records increased from 0% to 70%. At the end of the 27-day trial, final body mass, weight gain, weight gain rate, specific growth rate, average daily gain and condition factor did not differ significantly between the device and control groups (P>0.05). Mortality was 4.66% and 1.27%, respectively, and feed conversion ratios were 1.01 and 0.93, respectively. Plasma cortisol did not differ significantly between groups (P>0.05).

Conclusion Coordinated optimization of feeding-station structure and counting logic for the rapid feeding behavior of L. maculatus improves system adaptation to feeding events, and enables preliminary automatic identification of individual feed-intake under group-rearing conditions. The current accuracy of 70% requires further improvement, and device effects should be validated with greater experimental replication and larger physiological sample sizes.