Measurement of Morphology Parameters of Dispersed Media Based on Polarization Imaging

The convolutional neural network reconstruction technology based on polarization imaging offers several advantages, including non-contact operation, robust recognition, and high precision. This technology enables the extraction of microstructure information from dispersed media. In this study, the approach is employed to simulate the entire process of reconstructing morphology parameters for dispersed media. This involves establishing a polarization imaging model for non-spherical dispersion media, using a Monte Carlo simulation program to replicate the imaging process of dispersion media, creating a dataset, and training a convolutional neural network model on this dataset. Then the fitting results for both the training and testing sets are separately subjected to linear fitting. The results demonstrate the effectiveness of this technology in reconstructing morphology parameters based on the scattering characteristics of dispersion media.