Identification And Classification Of Subtle Defects On Metal Surface

The identification and classification of workpiece surface defects is a hot issue in the field of machine vision research.The detection of scratches,spots,pits,and other defects on the surface of the workpiece is critical to ensure product quality.This paper has conducted related research work based on this topic.The first is the problem of image preprocessing.In this paper,based on the type of noise in the acquired defect images,the neighborhood weighted average filter and the median filter algorithm are used to remove Gaussian white noise and salt and pepper noise in the image.For the subtle defect feature is not obvious enough,the Laplace operator is used to sharpen the image and enhance the defect features.The second is the extraction of defect features.For the extraction of subtle defect features in the non-texture background,based on the analysis of the original kernel principal component,the second-order oscillating particle swarm optimization algorithm is applied to optimized nuclear principal component analysis.An improved kernel principal component analysis algorithm is proposed to extract the defect features.Aiming at the extraction of subtle defect features in textured backgrounds,this paper proposes a defect feature extraction algorithm based on low rank matrix reconstruction.This algorithm models the background image of the metal surface texture into a low rank matrix,and models the foreground image with defects into a sparse matrix.Then using the accelerated proximal gradient method(APG)to solve low-rank matrix reconstruction problem.At the same time,in order to solve the difficult problem of segmentation coefficient selection,this paper proposes an iterative correlation method for progressive iterative approximation to select the best segmentation coefficient.The last is to classify the defects.According to the characteristics of metal surface defects,a hierarchical kernel clustering analysis algorithm based on support vector machine was proposed to classify the defects.The experimental results show that the improved kernel principal component analysis algorithm improves the defect classification accuracy compared with the original kernel principal component analysis algorithm;the low-rank matrix reconstruction algorithm can effectively separate the background texture and improve the defect classification accuracy;The classification accuracy of hierarchical kernel clustering analysis algorithm is higherthan that of support vector machine and K-means cluster analysis,and the accuracy rate is94.4%.