A Novel Quantitative Analysis Method for Studying Animal Behavior and its Interdisciplinary Application in the Study of Feeding Mechanisms

Abstract

Animal behavior research, long rooted in traditional biology, has evolved to intersect with various scientific disciplines. The integration of machine learning and computer vision represents a transformative development, capturing and analyzing immense data related to animal behavior, from simple movement observations to complex behavioral patterns and group dynamics. Convolutional Neural Networks (CNNs) exhibit remarkable precision in classifying and detecting behavioral changes over time, providing previously inaccessible insights.

Despite substantial progress, a comprehensive understanding of animal behavior and its underlying mechanisms presents unanswered questions. Key challenges include designing experiment setups allowing for free movement and accurate behavior recording, unexplored territory lies in subjective recognition of behavior segments through video analysis combined with non-video modalities, and the development of real-time animal stimulation methods for monitoring and interaction remains to be fully realized.

In this thesis, we propose a novel approach for the quantitative study of animal behavior, combining the power of machine learning and computer vision, focusing on computational ethology. We have developed a versatile behavioral analysis system that integrates multiple sensor inputs for high accuracy and fast recognition of mouse behavior. This system enables real-time monitoring and closed-loop manipulation of animal behavior in response to stimuli.

Moreover, we introduce DLATA (Deep Learning-Assisted Transformation Alignment), a method that accelerates brain slice alignment processes in neuroethology, aiding in accurately identifying and localizing neurons within specific brain regions.

To assess our approach, we conducted experiments on hungry mice during feeding time under various conditions to study feeding behavioral mechanisms. Results revealed significant behavioral differences between the experimental and control groups, identifying specific feeding patterns under diverse circumstances. The data collected and analyzed from these experiments illustrate the potential of our behavioral analysis system and DLATA for future animal behavior studies.

Overall, our research emphasizes the need for further interdisciplinary collaboration in animal behavior research. Our proposed approach offers a promising avenue for achieving a more accurate and comprehensive understanding of animal behavior and its underlying mechanisms.

Date of Award	8 Sept 2023
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Ho Man CHAN (Supervisor)