Particle positioning and characterization using wavefront curvature with a convolutional neural network in digital holography

Accurately and simultaneously determining the axial position, radius, and refractive index of spherical particles remains challenging in digital holography (DH). Applications such as fluid dynamics, fuel droplet evaporation, and aerosol characterization often require low-numerical-aperture (NA) optical setups to capture a broad field of view. However, the reduced number of interference fringes hinders accurate parameter estimation using conventional holographic interference pattern analysis. To address this limitation, we present a method that employs a one-dimensional convolutional neural network trained on wavefront curvature profiles along the optical axis. This approach enables the accurate and simultaneous determination of all three parameters for low- and high-NA systems. In a low-NA (0.02) setup, our method achieves accuracies of 0.3%, 2.0%, and 7.0% for radius, axial position, and refractive index estimation, respectively, outperforming conventional methods. Experimental validation confirmed the effectiveness of our approach, enhancing the capabilities of DH for particle characterization across diverse applications requiring any NA setup.