A Graded Feeding Strategy for Aquaculture Fish

In fishery industry, accurate feeding based on the fish’s appetite is crucial. The feeding demand of fish is influenced by their characteristics, the environment, and other factors. Thus, manual feeding cannot simultaneously account for these factors, limiting accuracy. In response to this question, this paper proposes detection network for residual fish feed and a four-level model for classifying fish’s feeding intensity. Then, a graded feeding strategy was developed by combining the two models. This strategy adjusts the feeder’s start and stop times based on the small fish’s appetite to achieve intelligent feeding. Results showed that the mean absolute error (MAE) and root mean square error (RMSE) of the feed counting network were 3.23 and 5.14, respectively, which greatly surpassed the state-of-the-art model, and the network demonstrated a large advantage in small-volume feed counting. The classification network achieved high accuracy in automatically classifying feeding intensities and justified the four-level classification’s reasonableness. Finally, a graded feeding strategy was developed by integrating the amount of residual feed and the intensity of small fish feeding activities. This strategy allowed for graded adjustments of feeding intervals and determination of feeding endpoints, which were compared to manually judged feeding endpoints to verify the strategy’s effectiveness and accuracy. The data is available at: https://github.com/pigcat88/Graded-feeding-strategy.