Research And Application Of Industrial Visual Inspection Method For Complex Surface Quality

The visual inspection technology of complex surface quality is more and more widely used in industrial production.In this paper,inspired by the behavior that inspection worker observe surface from different angles to obtain defects information,a multi-angle lighting multivariate image analysis approach was proposed.The experimental results of detecting metal surface defects verify the reliability of this method;In order to increase the generality for similar surface classification tasks,the color texture feature library was established.Feature selection method combined with Filter and Wrapper was used to grade various types of ceramic tile.The main research work of this paper is as follows:(1)In the application of metal surface defect detection,by adjusting the lighting height selectively,multivariate images could be constituted from different lighting angles.Multivariate images with effective lighting angles were stacked and unfolded,from which the Q-statistic image could be calculated based on the Multivariate Image Analysis(MIA)technique.(2)With the analysis of obvious score image and the corresponding loading vector,the effective lighting angle could be selected,for the reason that loading vector could reveal the influence weight of images with different channels to the obvious score image.In the experiment the channels number was reduced from 37 to 9.(3)By the block processing to the Q-statistic image,the problem of uneven illumination could be effective solved.Then the texture feature,obtained from the small image,was used to train the SVM classifier for defect detection.The results with high detection rate and 6.6% pseudo reject rate verify the effective of this method in metal surface defect detection.(4)In the application of ceramic tile grading,the feature selection method is quite necessary.With the analysis of the character of the feature selection based mutual information and the KNN classifier,the feature selection method combined with Filter and Wrapper was proposed.(5)The color texture feature library was established with 6 different features,which is 292 dimensions totally.Six types of ceramic tiles obtained from VxC TSG tile database was used to feature extraction,feature selection and grading test.The results shows that the overall classification accuracy was higher than 95%,which verify the effective of feature selection method in ceramic tile grading.