<p>Amid growing protein demands from population expansion, aquaculture requires efficient and sustainable production solutions. Traditional feeding methods often cause inefficiency and environmental pollution due to empirical judgments and poor adaptability. To address these issues, this research proposes an innovative intelligent feeding decision-making (Ifeed) for farmed fish, based on the similarity measure of density distribution. It analyzes the feeding behavior of fish schools in real time and calculates the trend of fish aggregation and dispersion to realize accurate feeding. The method implementation is as follows: First, background subtraction and image preprocessing are used to extract fish school foreground images from multiple input data. Subsequently, based on the estimated density distribution, the key points of the target fish schools are localized and divided into zones. Finally, full connection similarity measurement is applied to calculate the degree of fish aggregation. Time series information is integrated to visualize the aggregation trend, assisting in feeding decision-making. The experimental results show that our algorithm significantly outperforms CSRNet and CMTL on the same dataset. Compared to CSRNet, our method improves MAE by 21.05%, MSE by 14.47%, and reduces prediction time by 83.57%. Compared to CMTL, the improvements are 70.00% in MAE, 65.79% in MSE, and 80.90% in prediction time. Through in-depth analysis and measurement of fish feeding behavior, this research provides an innovative and efficient Ifeed strategy for aquaculture. It can effectively decrease feed waste and water pollution, thus realizing cost reduction and efficiency increase in actual production to achieve the target of accurate feeding.</p>