Research On Phase Extraction Method Based On Single Electronic Speckle Pattern Interferometry Fringe Image

Electronic speckle pattern interferometry(ESPI)is a real-time measurement technology that uses optical principles to detect defects on objects.The technology uses laser speckle as the carrier of the changed information of the measured object,and detects the state change of the object through the correlation fringes of the interference speckle field generated when the coherent light is irradiated on the surface of the object to be measured.It can realize non-contact and full-field measurement with high sensitivity and good real-time performance.When applying electronic speckle interferometry technology for non-destructive testing,the accuracy of phase estimation is very important for non-destructive testing,displacement measurement and deformation measurement.To overcome the limitations of traditional phase recovery methods,this thesis studies the phase analysis method based on a single electronic speckle pattern interferometry fringe image.Two phase recovery models are constructed to realize fullfield phase recovery.One is the digital phase-shifted phase recovery model,the other is the end-to-end deep learning phase recovery model.The major parts are as follows:(1)For collecting the defect pictures of the measured object and establish the ESPI fringe-phase data set,an electronic speckle interferometry system based on the Michelson-type shearing interference optical path was built,which combined with a piezoelectric ceramic device to realize the physical phase shift of the speckle field.(2)A phase recovery method based on digital phase shift is proposed.For an electronic speckle pattern interferometry fringe image,the moving direction and moving distance of the fringe during digital phase shift are obtained based on the calculation of the fringe direction and fringe width,and the digital phase shift fringe pattern is generated.Based on this,the wrapped phase image is calculated and the fullfield phase is finally obtained.(3)An end-to-end phase recovery method based on deep learning is proposed.To improve the generalization ability of the model and achieve end-to-end full-field phase recovery,the U-Net basic network has been improved combine with the characteristics of ESPI stripes.