To address hypoxia-induced fish mortality, this study proposes an improved algorithm (MSSYOLO) based on YOLOv10. The algorithm is designed to enable real-time and accurate monitoring of surfacing behavior, thereby optimizing aquaculture management. MSSYOLO integrates multi-scale convolution in Conv and C2f modules, a Squeeze-and-Excitation mechanism for improved feature representation, and a P2 detection layer for enhanced small-object detection, achieving superior multi-scale feature fusion and lightweight design. Experimental results demonstrate that MSSYOLO outperforms baseline YOLOv10 models. Furthermore, A strong correlation between surfacing fish numbers and dissolve oxygen concentration highlights surfacing as a hypoxia indicator. This algorithm provides robust technical support for the real-time regulation of aeration systems, effectively preventing hypoxia-induced fish mortality and reducing economic losses.