Optical Lens Flaw Classification Test Research

At present,the domestic lens flaw detection mainly adopts the traditional manual detection method,but the cost of manual detection is high,so there is a problem that the detection standard is not objective or uniform.Therefore,the realization of optical lens flaw detection and classification of automatic detection has become an irreversible trend.Optical lens flaw detection classification is the realization of the lens automated testing of the problems to be solved.How to translate the physical characteristics of a lens flaw into computer-recognizable data features.And through the effective description of the way to integrate the physical characteristics of the lens flaws,lens flaw classification accuracy and timeliness.The main problems in this paper are the feature extraction and feature-based classification and implementation of the disease.In this paper,the flaw image features analysis,feature information enhancement,boundary feature extraction,geometric feature extraction research.And use the extracted input features using SVM to learn and classify.Specifically,is to analyze the types of lens flaws and causes.According to histogram characteristics and geometric characteristics of different flaws,the characteristics of high contribution rate of classification are analyzed,and the classification strategy is given.According to the research of edge feature extraction,comparing several edge detection operators,it is found that the edge detection method can not obtain complete boundary information and is very sensitive to noise.Which can not get the exact geometric characteristics of the lens flaw.This paper presents an improved method of regional edge detection to improve the timeliness and accuracy of the boundary information extraction.In order to improve the accuracy and timeliness of the geometric feature calculation,an improved method of calculating the bounding rectangle of minimum area is given.The linear correlation is an important classification feature of scratches and flocks.However,if the width of the scratched pixel is too large,the linear correlation coefficient can not reflect the linearity of the scratches well.In this paper,the linear correlation coefficient is calculated after refinement.Aiming at the calculation of area and perimeter,an area and perimeter calculation method based on area filling coding is proposed.According to the extracted features,PCA principal component analysis(PCA)is used to reduce the dimension of eigenvectors,SVM is used to learn and classify,and a SVM model with high classification success rate is established.According to the analysis of the characteristics of the lens flaws,targeted extraction and processing,thereby enhancing the accuracy and timeliness of the lens detection system.