Research On Digital Holographic Microscopic Topography Measurement Technology Based On Deep Learning

With the rapid development of microelectromechanical systems,microorganisms and integrated circuits,the demand for measurement technologies such as cells,MEMS sensors and micro nano chips is increasing.Therefore,digital holographic microscopy has gradually become a research hotspot in the field of topography measurement.Digital holographic microscopy is a new optical three-dimensional imaging technology,which combines digital image processing to realize the function of observing and measuring the sample,and it can also quantitatively analyze the three-dimensional structure of the measured object.Therefore,after decades of development,digital holographic microscopy has become a widely used technology in sample inspection.At the same time,due to the advantages of digital holographic microscopy,such as label free,non-contact,real-time and non-invasive,it has broad prospects in the field of device measurement,microfluidic and nano particle tracking.The main research contents as follows:(1)According to the current research status of digital holographic microscopy,the imaging principle,three reconstruction algorithms and phase unwrapping algorithm of digital holographic microscopy are systematically studied in this paper.At the same time,due to the introduction of off-axis angle and microscope objective in the optical path of digital holographic microscope,the phase distortion of the unwrapping phase of the measured object is brought,which seriously affects the 3D reconstruction effect of the measured object.Therefore,the phase distortion compensation algorithm in digital holographic microscopy is studied in detail.(2)Because the commonly used numerical compensation method includes the phase value of the measured sample in the process of fitting the phase distortion,which has a great impact on the fitting results,so the error of the reconstruction results is large.Aiming at the limitation of current algorithms,this paper proposes a solution based on deep learning,and builds u-net convolutional neural network which is widely used in the field of image segmentation.The background region of the hologram is extracted by deep learning image segmentation technology,and the unwrapping phase map without the sample phase value is obtained.Then the background region without the tested sample is fitted,so as to eliminate the influence of the tested sample on the fitting.The proposed algorithm is verified by the simulation of the hologram.(3)Finally,based on Michelson interferometer,a digital holographic micro optical structure is designed,and a set of experimental platform is built to measure the processed samples.The reconstruction results of different phase distortion compensation algorithms are compared through experiments,and the height of the observed sample is measured by three-dimensional measurement laser microscope,which proves the feasibility and accuracy of the proposed algorithm.