Research progress of particle field digital holography based on deep learning

In the digital holography of particle field, reconstructing a high-precision three-dimensional particle field distribution from a particle field hologram is one of the classic problems in the field of digital holography. Compared with the traditional inverse reconstruction algorithm, the deep learning algorithm can directly reconstruct the three-dimensional particle field from a single hologram to simplify the algorithm complexity and improve the calculation efficiency and accuracy rate. The research progress of particle field digital holography in combining deep learning algorithm with digital holography technology by the research teams at home and abroad was introduced. Starting from different methods of particle characterization, the application principles, implementation approach and accuracy rate of the support vector machine, fully connected neural network, fully convolutional network, U-Net network and deep neural network in the process of particle characterization and particle field inverse reconstruction for particle field holography were described. Finally, the advantages and shortcomings of deep learning algorithm in this research field were pointed out, and how to further improve the accuracy of this method was prospected.