Frequency Spectrum Based Optimal Texture Window Size Selection For High Spatial Resolution Remote Sensing Image Analysis

The features on the high-resolution images have a high degree of recognizability,such as shapes,colors,boundaries,etc.are clearly visible.At the same time,there is a phenomenon that similar features have multiple spectral characteristics.It is difficult to achieve accuracy requirements by spectral information classification Therefore,texture features are often used as an effective complement to spectral features and are introduced into remote sensing classification.Whether it is traditional pixel classification,or later object-oriented classification,and now deep learning classification,scholars attach great importance to texture features,and the addition of texture features to classification can effectively improve the classification accuracy.However,in the process of texture feature expression,there is a very worthy concern,that is,the problem of texture scale,that is,the size of the window involved in the texture extraction process.Although there are more mature theoretical systems and research methods for scale effect analysis of remote sensing images,and texture scale analysis can be categorized as a small aspect of this large system,due to the complexity of texture features to express their own methods,texture scale selection And texture scale effect analysis still lacks an effective method.In view of the above problems,considering that the measurement of texture is more intuitive and convenient in the frequency domain than in the spatial domain,the energy map is obtained after the Fourier transform of the image,and the polar coordinate measurement can accurately describe the distribution of the periodic pattern and direction of the image texture.The most commonly used methods of texture feature expression,namely Gray Level Co-occurrence Matrix and fractal model,are the basis of texture feature expression method.Fourier spectrum statistical analysis method is used to analyze the scale effect of the above two texture features.The purpose is to determine the theoretical optimal texture scale that can achieve the best classification based on the characteristics of the texture itself,so that the image texture information can be more fully utilized.The main work and results of this article are as follows:(1)This paper systematically summarizes the research status of remote sensing scale issues,clarified the relationship between texture scale and remote sensing scale,and pointed out the position of texture scale in the remote sensing scale issues research system.This paper expounds the existing texture scale research,points out the problems existing in the existing research,and proposed the solutions according to the problems.(2)This article proposes the best texture scale selection method based on the spectrum statistics method.In the process of analyzing the changes of the spectrum energy of the image in the frequency domain under different scales,it clarifies how to choose the best scale by the spectrum analysis method.The essence of this change is caused by the changes of the texture period and direction.The features of ground objects,texture and the best scale are closely related,and the most appropriate texture scale not only keeps the period and direction of the original texture,but also has the lowest statistical value of energy.(3)Firstly obtain the ground features of different textures on the original remote sensing image and extract the multi-scale texture image,and then transform the multiscale texture extended image into the frequency domain,draw the radial and angular spectrum curve graphs of each texture at different scales,through analysis the radial and angular spectral energy change information at different scales is used to obtain the best expression scale of the texture.In order to prove the effectiveness of this method,this paper uses high-resolution image to carry out multi-scale classification based on a series of texture scale expressions,uses SVM to classify multi-scale images,and evaluates the accuracy of classification results.The experimental results show that using the best scale texture based on spectrum analysis for classification can effectively improve the quality and accuracy of image classification.