Research On Surface Defect Detection Technology Of Mobile Phone Metal Plate Based On Machine Vision

Due to many factors such as equipment and technology involved in the manufacturing process,the mobile phone metal plate surface inevitably has various defects.These defects have great influence on the wear resistance,corrosion resistance and electromagnetic properties of the metal plate.Based on the research status of the existing surface defect detection algorithms at home and abroad,the paper discusses the surface defects of metal plate with different types and shapes,small irregular distribution,insignificant gray contrast of defects,and combines image processing,computer vision and pattern recognition to study the key technologies of image denoising,defect feature description and classification and recognition of surface defect.The characteristics of metal plate surface defect put forward higher requirements on the preservation of image information in the process of denoising.Therefore,a non-local mean denoising method based on rotationally invariant and noise robust feature called FoPLBP is designed to denoise the surface defect image of the metal plate,and then the denoised surface defect image is chunked to obtain a data set of 1,725 defective images and 3,075 non-defective images.Given the phenomenon that the difference between the defect and the background of the metal plate surface image is not obvious,the local gradient feature HaarHOG is proposed,and the realization details of the HaarHOG feature extraction of the surface defect image block are given,and the classification experiment is carried out in the appropriate data set compared with the HOG feature.For the similarity between the minor defect and the noise existing on the surface image of the metal plate,only the HaarHOG feature cannot be used to accurately classify and the local frequency feature Gabor feature of the surface defect image block is extracted.For defect image block classification,firstly based on principal component analysis,the dimensionality reduction of high-dimensional defect feature is realized;then two classification models are selected of Naive Bayes and Support Vector Machine,and comparative experiments are conducted in different classification designs with the HOG+Gabor feature.The experimental comparison shows that the HaarHOG + Gabor feature is superior to the HOG + Gabor feature for the surface defect of the metal plate studied in this paper.The classification result in the PCA feature reduction + RBF kernel SVM meets the requirements of the detection accuracy.