Synergistic Enhancement of Detection-Tracking Framework for Zebrafish Shoaling Behavior Analysis

The integration of computer vision and artificial intelligence (AI) technologies has injected new momentum into computational biology. However, compared to individual animal behavior analysis with established standardized workflows, quantitative research on shoaling behavior has lagged, because of the lack of algorithmic frameworks for shoaling behavior quantification. To address this issue, this paper proposes a cascade-enhanced quantitative technical framework combining object detection and object tracking for shoaling behavior analysis, the framework achieved 98.22% tracking accuracy and 97.07% identification precision. And we extract a multidimensional feature set encompassing both kinetic and spatial distribution characteristics for shoaling behavior. Furthermore, by employing the proposed multidimensional feature set, we observed that ethanol exerts a biphasic modulation on the global motion intensity of zebrafish shoals, specifically, low-concentration ethanol induced hyperactivity, while high concentrations caused sedation. Additionally, ethanol dissolved the shoal structure and attenuated inter-individual communication.