Deep learning variable-density low-quality electronic speckle stripe direction extraction method

Abstract

The invention belongs to the technical field of optical detection and image processing, and relates to a deep learning variable-density low-quality electronic speckle stripe direction automatic acquisition method in order to realize batch full-automatic acquisition of a direction field for multiple low-quality variable-density ESPI stripe graphs, which comprises the following steps: 1, respectively constructing a training data set and a verification data set; 2, constructing a network model for calculating low-quality and variable-density electronic speckle stripe directions; 3, optimizing parameters in the back propagation process through a gradient descent method to perform model training; and 4, inputting the verification data set into the network, training the network model by using the training data set to obtain a model, verifying the model, testing the generalization ability of the model, and finally automatically obtaining the stripe direction by using the trained and verifiedmodel. The method is mainly applied to optical detection and image processing occasions.

Description

technical field [0001] The invention belongs to the technical field of optical detection and image processing, and relates to a batch automatic direction calculation method of electronic speckle interference fringes based on deep learning. Background technique [0002] Electronic speckle interferometry technology, as a modern optical measurement method developed on the basis of modern high-tech achievements, has the advantages of high accuracy, rapid response, and non-contact throughout the field. It has achieved good results in the measurement of three-dimensional surface topography of objects, the measurement of mechanical properties of materials, the measurement of in-plane displacement, vibration monitoring, and the measurement of thermal deformation of test pieces. However, there is strong granular noise in the speckle fringe image, which greatly limits the visibility and resolution of the fringe. In addition, the noise spectrum and the fringe spectrum are mixed togethe...

AI Technical Summary

Problems solved by technology

Although there are many traditional methods for calculating the direction of texture images, for calculating the direction of low-quality speckle images with low signal-to-noise ratio, variable density, poor contrast, and fine stripes, the parameter adjustment is complicated, and the calculation results need to be improved.

Images