Multi-view image-based behavior classification of wet-dog shake in Kainate rat model

Area of Science:

• Neuroscience
• Animal Behavior Analysis
• Computer Vision

Background:

• Wet-dog shake behavior (WDS) is a significant indicator in animal models for conditions like seizures and substance withdrawal.
• Existing animal behavior detection systems do not capture WDS due to its short duration and specific characteristics.
• Accurate WDS detection is crucial for advancing research in various neurological and addiction models.

Purpose of the Study:

• To introduce a novel multi-view animal behavior detection system capable of identifying WDS.
• To evaluate the system's performance using image classification and a time-multi-view fusion scheme.
• To demonstrate the system's adaptability to different animals and behaviors without manual feature engineering.

Main Methods:

• Developed a multi-view animal behavior detection system utilizing image classification.
• Implemented a novel time-multi-view fusion scheme for behavior analysis.
• Tested the system's efficacy in detecting WDS in rats using varying numbers of camera views.

Main Results:

• The multi-view system successfully detected WDS behavior in rats.
• Performance analysis showed that increased camera views significantly improved WDS classification accuracy.
• With three cameras, the system achieved a precision of 0.91 and a recall of 0.86.

Conclusions:

• The presented multi-view system is the first to successfully detect wet-dog shake behavior (WDS).
• The system's performance improves with the addition of multiple camera views, highlighting the benefit of multi-view fusion.
• This technology offers significant potential for applications in diverse animal disease models and behavior research.