A convolutional neural network technique for online tracking of the radius evolution of levitating evaporating microdroplets of pure liquids, liquid mixtures and suspensions
Description
In this study, we use a convolutional neural network ( which we trained on theoretically generated Mie scattering patterns, grouped into classes encompassing consecutively small ranges of radii) to classify corresponding experimentally recorded sequences of patterns, enabling the tracking of the radius evolution of evaporating microdroplets of pure liquids, liquid-liquid mixtures, and suspensions. The zip file contains 3 folders: "Convolutional Neural Networks" which contains the MATLAB script for the untrained convolutional neural network architecture, (t7g24d_Modified.m), the trained network (Stage1_Network_Pure_DEG.mat) and its corresponding class information. The class information is a ".mat" file containing 3 sub-vectors labelled "class" (class labels), "Range" (class radii ranges), and "Radius" (class average radii). Finally, the folder contains a MATLAB script "Classify_images.m" for classifying the experimental images sequentially. The 2 other folders contain images (generated for 3 classes and experimental images. The full length article can be accessed from Journal of Quantitative Spectroscopy and Radiative Transfer: https://doi.org/10.1016/j.jqsrt.2025.109533 Or from arxiv : https://doi.org/10.48550/arXiv.2410.08857
Files
Institutions
- Polska Akademia Nauk Instytut Fizyki
- University of Cape Coast