Texture Classification And SAR Image Recognition Based Graph Wavelet Transform

Texture is the intrinsic property of an object,which can be represented in the form of patterns,shapes,roughness and so on.Due to its universality,the texture,as a basic research unit in image processing,plays an important role in many practical image applications,including radar image processing,medical image analysis,forest species identification,etc.Worked as an application extension based on texture analysis,SAR image recognition has been widely used in many practical fields such as military and national defense.Most of the previous texture analysis methods are based on wavelet transform.Compared with the traditional wavelet transform,the graph signal processing can clearly consider the structure and relationship between signal samples while retaining its multi-scale analysis and noise reduction characteristics,so it is more suitable for texture analysis than the traditional wavelet transform.In this paper,the texture analysis method based on graph wavelet transform is used to complete image texture classification and SAR image recognition.Its main research contents include the following two aspects:(1)An image texture classification method based on graph wavelet transform framework is proposed.The texture image is decomposed into multi-scale components by using wavelet filters.Considering that the traditional singular value decomposition has fewer coefficients,it is impossible to estimate the model well.In this study,the sub-bands obtained by wavelet transform are used to perform local singular value decomposition.In order to be robust to noise,the maximum,average and median values of the local singular values are extracted in the experiment.Then,the elements in each set are respectively modeled as Weibull distribution to describe the image texture,and the KL divergence of the two distribution functions is used as the sample distance to classify the texture samples.This method uses three data sets constructed by two databases to complete the experiment,and the experimental results demonstrate the effectiveness of the proposed scheme.(2)A method for target recognition of SAR images using the above texture analysis framework is proposed.Since SAR image recognition belongs to surface target recognition in target detection,texture analysis is still the decisive factor in radar image recognition.This method provides a multi-scale analysis by using the wavelet transform for SAR images.In view of the messy background and the strong speckle noise,the two-dimensional principal component analysis(2DPCA)method is used to suppress the image noise while extracting the main features of the image.Then,we use the weighted sparse representation classifier to perform target recognition for SAR images,and improve the existing method to calculate the weight distance during classification.The correlation distance between samples is adopted to calculate the weights needed in sparse representation.Finally,three types and ten types of SAR images in the MSTAR database were used in this experiments.Compared with the existing schemes,the experimental results proved the superiority of this scheme.