Structure-Aware Multi-Animal Pose Estimation for Space Model Organism Behavior Analysis.

Area of Science:

• Space biology
• Animal behavior analysis
• Biotechnology

Background:

• Multi-animal pose estimation is vital for quantifying group behaviors under space environmental factors.
• Space biology experiments on the China Space Station use model organisms (C. elegans, zebrafish, Drosophila).
• Existing methods struggle with species-specific differences, challenging generalization and robustness.

Purpose of the Study:

• To develop a flexible and general single-stage multi-animal pose estimation method.
• To address challenges posed by diverse species types, body scales, and posture dynamics in space environments.
• To provide robust pose estimation for model organisms in space biology research.

Main Methods:

• Proposed a novel single-stage multi-animal pose estimation method.
• Constructed species-specific pose group representations using anatomical priors.
• Incorporated multi-scale feature sampling and structure-guided learning for enhanced robustness.

Main Results:

• Evaluated on the SpaceAnimal dataset, the first public benchmark for space organism pose estimation.
• Achieved superior AP scores: 72.8% for C. elegans, 62.1% for zebrafish, and 67.1% for Drosophila.
• Demonstrated effectiveness and robustness across different species and imaging conditions.

Conclusions:

• The proposed method offers strong technical support for on-orbit behavior modeling.
• Enables large-scale quantitative analysis of animal behavior in space.
• Advances the field of multi-animal pose estimation for space biology applications.