Deep Learning-Based Antibubble Dynamic Tracking and Behavior Analysis

Antibubbles are distinctive gas-enclosed liquid cores suspended in a liquid medium, forming a gas–liquid interface system. Their unique structural characteristics have attracted significant interest in applications such as microfluidic manipulation and drug delivery. However, during motion, they exhibit significant deformation, frequent occlusion, and complex interbubble interactions, posing challenges to high-precision tracking and quantitative analysis of their dynamic behavior. In this study, we construct and validate an analytical system for investigating antibubble dynamic behavior based on the DeepSORT multiobject tracking framework. By integrating a high-performance detection module with the DeepSORT tracking algorithm, the system achieves robust data association of antibubble motion trajectories and enables continuous monitoring throughout their entire lifecycle. Experimental results demonstrate that the proposed tracking system achieves a multi-object tracking accuracy of 90.5% and a multi-object tracking precision of 90.7%, significantly outperforming the traditional SORT algorithm. More importantly, the system enables automatic and accurate extraction of key morphological parameters and motion trajectories of antibubbles, with the relative errors of the measured parameters consistently below 4%. This study not only validates the effectiveness of deep learning-based tracking techniques for investigating complex fluid microstructures but also provides a novel automated analytical framework for the study of antibubble dynamics.