Research On Coating Defect Detection System Based On Machine Vision

In recent years,with the rapid development of graphic image technology and computer software technology,as well as the increasing quality requirements of consumers,a new technology-machine vision detection technology comes into being.In contrast,the traditional product defect detection is generally based on statistical theory,using the first sample after the artificial visual method,That is,from the bulk of the product to extract a certain percentage of products by piece to compare the difference between the production of products and standard samples,if the difference is acceptable within the range of qualified products,otherwise the defective goods.Practice shows that this method is not only labor intensity,and the detection accuracy can't guaranteed.Based on this,this paper studies the coating defect detection technology based on machine vision.This article first briefly introduces the basic hardware involved in this system.Then we extract the histogram features of the filtered preprocessed image and use the sigmoid function to compress the eigenvector,and the eigenvector matrix is obtained.Then a typical classifier-the nearest neighbor(k-nearest neighbor,knn)algorithm classifier is trained to classify the typical coating defects.Since the coated image with subtle defects may be misinterpreted into defect-free images,the principal component analysis(pca)algorithm continues to implement detection,identify and locate the defects.In order to achieve the real-time requirement,the algorithms involved in this system are deeply optimized using MKL(Math Kernel Library).The results of the follow-up experiments show that the above-mentioned algorithm has obvious effect on the classification of typical coating defects and the detection of fine defects by reasonably setting the parameter values involved in the algorithm.At the same time,thanks to the bottom-level optimization of MKL algorithm,the real-time performance of this algorithm can meet the production requirements.