Recognition Method Of Metal Fracture Images Based On Empirical Ridgelet And Principal Component Analysis

The accurate extraction of fracture image features and the effective compression of feature space are of great significance for the identification of fracture images.Based on this,this paper introduces the empirical Ridgelet to the feature extraction of fracture images and uses the principal component analysis method to compress the feature space.The fracture image recognition method based on empirical Ridgelet and principal component analysis methods is proposed,and good innovation results are obtained.The main research content of the paper is as follows1.Introduced the theory and the algorithm of Empirical Ridgelet transform,and combining the advantages of PCA in feature space compression,an image recognition method based on empirical Ridgelet-PCA is proposed.The average energy,entropy,and kurtosis of the empirical Ridgelet coefficients are used as characteristic parameters to describe the fracture images.Several characteristic parameters are compressed by principal component analysis(PCA)to obtain the main component of each characteristic parameter.Finally,the comparison analysis between the experiment and the Ridgelet-PCA fracture identification method was conducted.Through experiments,it is found that the basis function of the empirical Ridgelet transform can be changed according to the content of the image when the image is decomposed,while the Ridgelet transform does not have this feature.2.For the problem that the PCA lacks the capability of extracting the image's nonlinear feature information,an image recognition method based on empirical Ridgelet-KPCA is proposed.Experimental results show that the empirical Ridgelet kurtosis-KPCA recognition results are optimal.The empirical Ridgelet kurtosis is the fourth-order relationship of empirical Ridgelet coefficients and belongs to the high-order nonlinear characteristic information of fracture images.Therefore,the proposed Ridgelet-KPCA identification method can effectively extract the nonlinear characteristic information of fracture images.3.The empirical Ridgelet transform can decompose an image into a set of independent BIMF components.The BIMF components contain local feature information of different frequencies and different spaces of the original image.The optimal BIMF component of the metal fracture image is extracted by introducing the calculation of entropy of the empirical Ridgelet coefficients,and the recognition method of fracture image based on the empirical Ridgelet-2DPCA is proposed.The experimental results show that the recognition rate of empirical Ridgelet-2DPCA is higher than that of empirical Ridgelet-PCA.This is mainly because the sample size of 2DPCA is small,and the estimation of the image's covariance matrix is more accurate.In addition,when feature space compression is performed,the selection of feature space dimensions also has an important influence on the recognition results.When the dimension of selected feature space is small,the feature information of the image is not perfect,and the image recognition rate is low;when the dimension of feature space selected is too large,the feature information of the image is redundant,which will increase the computational complexity,and may also reduce the recognition rate of the image.4.For the problem that 2DPCA can only extract the principal components in a single direction of the image,the bilateral two-dimensional principal component analysis is introduced into the metal fracture image recognition,and the fracture image recognition method of the empirical Ridgelet-B2 DPCA is proposed.Experimental results show that B2 DPCA can extract principal components from both the row and column directions of the BIMF component,and the size of the resulting feature matrix is much smaller than that obtained from 2DPCA.Therefore,the empirical Ridgelet-B2 DPCA algorithm not only has excellent recognition results.For experience Ridgelet-2DPCA recognition algorithm,and recognition time is much less than the experience Ridgelet-2DPCA recognition algorithm.