Research And Application Of Metal Surface Defect Detection Algorithm

Sheet metal is one of the important materials in industrial production, so improving thequality of sheet metal is considered as the main means to improve the enterprise’s marketcompetitiveness,The existing surface defect detection system has been unable to meet theneeds of sheet metal production.In this topic, several common defects images are taken as the research objects.A researchof the images smoothing, the image enhancement of the metal surface defects, the imagesegmentation of the metal surface defects and classification recognition of metal surfacedefects has been done.Firstly,considering the defect images which exsit some extra noise needs to removed bythe processing of defect image.Comparing three kinds of traditional filtering algorithm,thispaper proposed an improved adaptive median filter, which combines the mean filter andmedian filter, to make the basic image smoothing,getiing rid of the extra noise existing in theimages.The improved adaptive filter can not only remove image noise,but also retain imagedetails and edge information better.Secondly,in the research about the wavelet transform theroy,considering most of thelength of the wavelet filters is greater than1,the length of decomposed image is finite,whichdirectly cause the distortion of the image reconstruction and biorthogonal wavelet just solvedthis problem.This paper conduct decomposition and reconstruction for metal surface defectimage based on biothogonal Mallat algorithm.Taken the entropy of each decomposed subgraph into account,the defect image is decomposed into2layers,then the sub images arereconstructed.The experimental results confirm that the proposed biothogonal Mallatalgorithm preserves approximate information of image,enhance the image information whileit reduces the processing of the image data.Thirdly,reconstructed images are segmented to extract the defects features.This paper useimproved mathematical morphology methods to process the reconstructed image, andconstructed3different structural elements to detect the edge of defect image,comparing withthe traditional edge detection operator, this method can detect the edge of defect better.Finally,Support Vector Machine was used to classify the defect images.The RBF kernelfunction are taken as the function of Support Vector Machine to find out the optimalparameter of RBF kernel function through the grid search and cross validation method.The experimental results show that, the biorthogonal Mallat algorithm proposed in thispaper is good for processing the surface defect images of metal, and the correct identificationof defect detection rate is about93%. Although this paper got certain progress, it still hassome gaps in online real-time automatic detection compared with other ideas, at the last of this paper some suggestions are put forwards for future work.