Shape reconstruction and size measurement of spherical particles based on single particle interference imaging and deep learning.

An algorithm for shape reconstruction and size measurement of spherical particles is proposed by combining interferometric particle imaging (IPI) with deep learning. Simulated and experimental interferometric defocus images of spherical particles are obtained from the optical transfer matrix theory and the IPI system. The Respe-Unet++, which adds Residual blocks (Res) and the Patch Expand (PE) module to U-net++, is proposed to reconstruct images containing shape and size information. The method is validated through simulation and experiments. The results indicate that Respe-Unet++ achieves a relative error of 0.00278% and a standard deviation of 0.071 in particle size measurement, with a measurement speed of 53.38 FPS. The analysis of incomplete images shows a relative error of 0.13% at a ratio of 10%. Compared to other U-net-based architectures, the Respe-Unet++ demonstrates superior performance in size measurement.