Research On Inversion Of Surface Defects Of Optical Components Based On Scattering Distribution

In recent years,the development of optical technology is changing rapidly.Optical systems composed of optical elements have been widely used in various fields such as scientific research,national defense and military affairs,and daily life.Especially in the field of national defense and military affairs,they have played an indispensable role.However,in the process of optical element processing,it is inevitable to induce surface defects.These defects seemingly insignificant will affect the stability and performance of the entire optical system seriously.So it is necessary to further improve the defect detection methods to meet the increasingly strict digital,objective and accurate requirements for defect detection.Therefore,the research about the surface defect detection of optical components is particularly meaningful.The paper is based on the scattering theory of surface defects on the optical components and the deep learning theory.The computer intelligence is introduced into the field of optics.The inversion of surface defects is combined with the task of image detection and recognition based on deep learning.The relationship between the defects and the scattering distribution is studied,the scattering distribution map is collected and the data set is made,and then,a deep learning network model suitable for the surface defect data set is designed and built.The model has been trained and tested and analyzed to realize the detection of defects.The inversion of scattering distribution to surface defects is completed.First,the theory of surface defect scattering and the angle-resolved scattering measurement method are studied,and the relationship between surface defects and scattering distribution is established.The deep learning method is introduced into the identification and classification of the scattering distribution map to judge the surface defect information.An inversion scheme for surface defects of optical components based on deep learning is established.Secondly,the data set of defect scattering distribution map is made to provide a data source for defect detection using deep learning methods.By using two methods:laser etching and single point diamond,non-defective samples and samples with scratches and pitting defects in series sizes are prepared.Scattering distribution maps of the samples are collected by the scattering measurement device.Multiple methods of data enhancement to enlarge the sample are used.Image preprocessing is performed on the scattering distribution map,and the production of the data set is completed.Finally,ResNet is the classic model of convolutional neural network,it is selected to train and test the sample data set.In order to further improve the problem of ResNet’s low accuracy on the data set,a model based on multi-channel features to enhance the residual network MCresnet is proposed.The detection effects of MCresnet model and ResNet model are compared and analyzed.Experiments show that the accuracy of the model based on the multi-channel feature-enhanced residual network has been greatly improved,and the accuracy of 95.81%,94.72%and 91.83%have been achieved respectively.It realizes defect detection and provides a reference for the optimization of subsequent models.In summary,deep learning is introduced into defect the detection,which realizes the inversion of defects from the three aspects of defect existence,defect type and defect size.Experiments show that the model’s ability to extract features from sample data is enhanced by the multi-channel feature-enhanced residual network.It proves the feasibility and effectiveness of using deep learning tools to the inversion of the surface defects based on the scattering distribution.