A convolutional neural network technique for online tracking of the radius evolution of levitating evaporating microdroplets of pure liquids, liquid mixtures and suspensions

• Evaporation of microdroplets: droplet radius being primary parameter of interest • A new technique is proposed using machine learning in a classification task • Theoretical light scattering patterns were generated from Mie theory • Convolutional neural network was trained on theoretical light scattering pattens • The trained network classified experimental images from evaporating microdroplets We have used convolutional neural network in a classification task to track the radius evolution of levitating evaporating microdroplets of pure diethylene glycol, diethylene glycol-water-polystyrene microparticles suspension, dipropylene glycol-water mixture and dipropylene glycol-water-silica nanoparticles suspension. We discretized a wider radii range into short radii segments, labeled them with class numbers and generated theoretically light scattering patterns from Mie theory. Then the network was trained on the theoretical images and used to classify unlabeled experimentally recorded Mie scattering patterns from the evaporating microdroplets. A plot of the class-average-radii versus the camera’s time step revealed the profile of the entire droplet radius evolution. We were able to work with ∼1500 classes and showed that the technique has the potential to distinguish droplet size difference of ±5 nm. We expect it to be applicable for online/real-time tracking of droplet evaporation.