Abstract: To achieve precise and adaptive feeding control in recirculating aquaculture systems (RAS) and solve problems such as low feed utilization and coarse growth regulation caused by static feeding strategies, this study proposes an intelligent feeding method integrating visual perception with Deep Q-Network (DQN). Taking freshwater grouper (Cichlasoma managuense) as the subject, real-time tracking of fish movement velocity is achieved through YOLOv8 and DeepSORT, and a quantitative index of fish school feeding intensity is constructed combined with texture features extracted from the gray-level co-occurrence matrix; furthermore, feeding intensity, water temperature, dissolved oxygen levels, and expected feeding intensity are taken as state inputs. A multi-objective reward function is designed, and a decision model is trained by Deep Q-Network to form a closed-loop control system. Experimental results demonstrate an average mAP@.5 of 85.3%. Under conditions of only 378.4 grams of average total feeding amount per fish, the decision model increased the fish weight gain rate (WGR) to 54.38% and reduced the feed conversion ratio (FCR) to 1.09, significantly outperforming traditional feeding methods. This approach effectively achieves real-time perception of feeding behavior and dynamic optimization of feeding strategies, providing a reliable technical pathway for precision management in recirculating aquaculture systems.