Machine learning techniques for positioning and characterization of particles in digital holography using the whole phase curvature

Abstract

We propose a machine learning-based regression method with the whole phase curvature of a reconstructed wave along the optical axis as input data to obtain not only the precise axial position but also the radius and refractive index of particles. Experimental results using well-characterized particles showed that an axial position of a particle could be detected, with the mean signed deviation (MSD) and root mean squared error (RMSE) being 0.02% and 85% of the particle’s diameter, respectively. A radius of 29.3 ± 0.36 µm and a refractive index of 1.589 ± 0.002 agreed well with the manufacturer’s specifications. In comparison to our previous nonlinear optimization method, the method was validated for characterizing the distribution of particle characteristics and can be used with a factor of 10,000 faster calculations.