Investigation of droplet deformation and breakup dynamics in porous media using machine learning-assisted computer vision

• ML-assisted computer vision improves parameter extraction for predicting breakup • Both droplet deformation and pair-wise interactions determine breakup probability • A normalized cut-off distance is a necessary but not sufficient condition for breakup • Framework enables robust prediction for optimizing droplet-based microfluidics Understanding and accurately predicting the deformation and breakup dynamics of emulsion droplets flowing through a porous medium is crucial for optimizing processes in enhanced oil recovery, filtration, and microfluidic technologies. A major challenge in these applications is the ability to precisely control droplet transport through complex pore constrictions, where multi-droplet interactions critically govern deformation and breakup behavior. Current tools lack the capability to quantify such transport and interactions across interconnected constriction networks, leaving collective dynamics in porous media poorly understood. Here, we present a data-driven framework that integrates high-speed imaging, machine learning-assisted computer vision, and statistical control-volume analysis to systematically quantify multi-droplet interactions and their influence on breakup dynamics in porous microchannels. This automated pipeline enables tracking and analysis of tens of thousands of droplets with high temporal resolution, extracting quantitative descriptors of size, velocity, deformation, and interaction patterns to capture population-level dynamics beyond traditional single-constriction or pairwise analyses. Through statistical modeling, we establish a necessary, but not sufficient, probabilistic criterion based on a normalized cut-off distance between droplets, demonstrating that pair-wise interactions significantly influence breakup likelihood in concentrated emulsions. This work provides a generalizable methodology for quantifying complex emulsion dynamics, offering a robust predictive framework for optimizing droplet-based microfluidic applications.