Font Size: a A A

Research On The Method Of Optics Damage Classification And Light Field Diffraction Ring Recognition

Posted on:2022-06-17Degree:MasterType:Thesis
Country:ChinaCandidate:Y W WuFull Text:PDF
GTID:2480306572950189Subject:Instrument Science and Technology
Abstract/Summary:PDF Full Text Request
The optical components in the high-power laser device will be laser-induced damage under high laser flux,which directly affects the performance of the entire device.Therefore,the damage of the optical components is detected in real time and the damaged optical components are repaired or replaced in time.The safe operation of the device is of paramount importance.The dark-field image in the final optics damage inspection(FODI)system uses sideillumination technology,and stray light is also collected when the damaged point is acquired.Stray light will cause false damage in the collected dark-field image,and affect the accurate determination of the damage location of the optical element.When the laser passes through the damage point on the optical element,it will produce diffraction.Since the damaged point to be detected is very small,it is approximately degenerated into a circular scattering source,and the generated diffraction fringes are mainly circular diffraction rings.Diffraction rings are easier to focus laser energy than other diffraction fringes,and are more likely to cause new damage to optical components.By detecting the diffraction rings in the light field,the growth process of damage can be found and tracked in time,which is of great significance for ensuring the performance of laser devices.In this thesis,the problem of authenticity and damage classification in dark-field images collected by online detection and the diffraction ring in near-field images are studied.The main research work is as follows:(1)The damage mechanism of the optical element is studied,the type of damage of the optical element and the cause of the induced damage are analyzed;the online detection method of optical element damage is introduced,and the cause of the diffraction ring in the FODI image is analyzed,which is set for the follow-up research theoretical basis.(2)Aiming at the problem of authenticity damage classification in dark-field images,several types of pseudo-damaged points in dark-field images that affect the determination of real damaged points are studied.The method of selecting 12 types of features combined with XGBoost classifier is used,and the parameters are carried out.Optimized to realize the classification of true damage,diffraction ring and mirror frame reflection type pseudo damage in dark field images,and the classification accuracy rate reaches 99.01%.(3)Aiming at the problem that the background of the light field diffraction ring in the nearfield image collected by online detection is cluttered and the scale changes are large,it is difficult to identify.Inspired by the histogram of directional gradient(HOG)algorithm,the HOG operator is combined with the simulated light field diffraction ring image to prove HOG The feasibility of the algorithm for the detection of diffraction rings;the method of HOG combined with support vector machine(SVM)is used to study the identification method of the light field diffraction ring in the actual near-field image,the parameters of the model are optimized,and the optimized model is used to perform the selected image Sliding window detection and real image detection experiments verify the feasibility of the HOG+SVM algorithm for detection of diffraction rings.(4)In order to deal with the problem of background clutter and large scale changes,and improve generalization ability,pyramid histogram of oriented gradients(PHOG)+ Mobile Net feature combined with SVM classifier is proposed as a near-field image diffraction ring recognition method,and PHOG selected as the feature,Combined with the lightweight neural network model Mobile Net,tested with actual images,and finally obtained the recognition accuracy of the diffraction ring at 95.44%,which is better than the result of HOG+SVM.
Keywords/Search Tags:Optical component damage detection, dark field imaging, Neural network, Machine learning, PHOG
PDF Full Text Request
Related items