Research On Fracture Image Processing Method Based On Empirical Curvelet Transform

This dissertation was supported by the National Natural Science Foundation of China(51675258,51261024),Science and Technology Projects of Education Department of Jiangxi Province,China(No.GJJ150699),National key research and development plan project(2016YFF0203000).Based on the deficiencies of existing metal fracture image processing methods,this project proposes a novel and adaptive signal processing algorithm,and then proposes a series of expansion methods for it.Then these methods are applied to the metal fracture image processing.The simulation and experimental results all show that the proposed method has certain advantages and performances.At the same time,these results are very innovative.The main content of this article is as follows:1.Based on metal fracture image processing,this article discusses its proposed background and its research significance.Metal fracture image processing includes image preprocessing,image feature extraction and image recognition.Based on this,this article gave a detailed overview of their research status at home and abroad.Finally,the research content of this paper was proposed.And the novelty and innovation of the method were proposed in this paper.2.The two-dimensional empirical curvelet algorithm was mainly introduced,and its unique advantages were verified through simulation studies.In this paper,we first analyzed the construction process of the first-generation and second-generation empirical curvelet transform algorithms.In terms of image processing,they used the advantages of empirical mode decomposition and curvelet transform.Therefore,they could not only represent the image sparsely,but also could construct different basis functions adaptively according to the geometric characteristics of the image itself.A filter bank was formed by pseudo-polar Fourier boundary detection.The basis function frame was constructed by it,which solved a series of problems such as wavelet transform and curvelet transform.In this paper,the core content of this algorithm,the two methods of Fourier boundary detection and their processes,were introduced and compared with the detection process of two-dimensional empirical wavelet 2D-EWT.Simulation was based on the detection method of scale space.The results showed the superiority of the 2D-EWTC method,which decomposed more ideal low-frequency information and more directional high-frequency components,and had great advantages for expressing singularity information such as edges,textures,and fine structures of an image.It was more conducive to image recovery,such as image denoising and reconstruction.This chapter provided a favorable theoretical basis for this article.3.Combining empirical mode decomposition and the curvelet transform in image denoising,this chapter put forward the second-generation empirical curvelet algorithm(2D-EWTC).In the standard image Lena and Barbara,Gaussian white noises with different intensities were added.Compared with The empirical wavelet transform and the threshold denoising methods of the first-and second-generation curvelet transforms,the proposed algorithm was superior in simulation.Finally,this method was applied to three kinds of metal fracture images with different textures.Under a series of noise levels,image denoising was performed based on four methods,and a comparative analysis was performed.The results show that the 2D EWTC method can completely recover and reconstruct the edge information,flat region information,contour and fine structure of the three metal fracture texture images.And it had certain robustness in image denoising.Finally,the advantages of this method were analyzed from the perspective of software runtime and speed.4.Combining 2D-EWTC and the concept of energy,entropy and kurtosis,three characteristic parameters,i.e.the energy of 2D-EWTC,the entropy of 2D-EWTC,the kurtosis of 2D-EWTC were proposed for feature extraction of metal fracture images.On this foundation,combining WLS-TSVM and all these parameters,WLS-TSVM recognition method based on 2D-EWTC was proposed.The proposed method was compared with SVM recognition algorithm based on 2D-EWTC and WLS-TSVM recognition method based on 2D-EWT in this chapter.The proposed method can remove the effect of classification hyperplane by setting penalty parameters and reduce the computational complexity by improving constraint terms compared with SVM recognition algorithm based on 2D-EWTC.compared with WLS-TSVM recognition method based on 2D-EWT,this method could extract more texture information in the metal fracture images.Experimental results showed that the proposed algorithm was superior to 2D-EWT WLS-SVM identification method and 2D-EWTC SVM identification method.