Abstract
Insect behavior refers to the observable actions and activity patterns of insects in response to internal physiological states and external environmental stimuli. Understanding and quantifying insect behavior can provide essential insights in many applications, including agriculture, ecology, neuroscience, and bioengineering. Although current computer vision and deep learning technologies have achieved remarkable progress in insect detection and recognition, their application in the quantitative analysis of insect behavior remains relatively limited. This article provides a comprehensive review of computer vision and deep learning technologies for insect behavior analysis, spanning application domains, research organisms, behavior categories, and adopted methods. The survey summarizes the methodological frameworks into spatial part location methods and spatiotemporal feature extraction methods, systematically examining their key components from video acquisition, data annotation, object tracking, to behavior quantification. In addition, representative approaches are analyzed and compared in terms of their strengths, limitations, and applicability under different experimental conditions. The survey also scrutinizes extant datasets and open-source platforms, offering a detailed explanation of dataset characteristics, annotation challenges, and factors affecting model generalizability. Finally, emerging research directions, including foundation models, self-supervised learning, and multimodal sensing for behavior analysis, are discussed to highlight potential opportunities for advancing the field of insect behavior studies.