High Resolution Remote Sensing Imagery Change Detection Based On Random Forest And SVDD Feature Selection

In recent years,high-resolution remote sensing image production technology has been continuously improved,and both high-resolution satellite imagery and aerial image data acquisition have become increasingly convenient.Remote sensing image change detection technology is one of the many applications of remote sensing images.With the continuous increase of data resources,its technical requirements are constantly updated.As a new image analysis paradigm,object-oriented technology provides a new idea for the detection of high-resolution remote sensing image change detection.On the basis of object-oriented analysis,a large number of remote sensing image features can be extracted and this series of features can be added to the image analysis,which can greatly enhance the accuracy of image change detection.At the same time,as the method of machine learning continues to be applied in so many research fields,its application in the detection of changes in remote sensing images has yet to be explored.By studying the basic theory of object-based remote sensing image change detection and machine learning classification and feature selection techniques,this paper studies the random forest classification algorithm for high-resolution remote sensing image change detection,and proposes a corresponding improved algorithm framework.The main work can be summarized as follows:(1)Explains the key technical issues of high-resolution remote sensing image change detection in detail,and conduct in-depth research on the characteristics of remote sensing images based on objects.According to the principles of change detection technology and the characteristics of remote sensing image objects,the spatial neighborhood characteristics of high-resolution remote sensing image objects are explored and extended,and a spatial neighborhood feature measurement method between objects and objects is proposed and applied in the entire change detection experiment.(2)The principle of SVDD(support vector description)feature selection and random forest classification algorithm is systematically explained.Aiming at the application features of classification technology in direct remote sensing image change detection technology,the research of feature selection using single classification technology was attempted.Two object-oriented change detection algorithm frameworks for high-resolution remote sensing images based on SVDD feature selection and random forest classification algorithms are proposed.(3)For the traditional manual selection of samples based on classification algorithms,this paper uses a classification framework for automatic sample selection based on CST(Chi-Square Transform)feature fusion and EM(Expectation Maximization Algorithm)automatic threshold segmentation techniques and applies to the proposed two SVDD feature selection and random forest classification frameworks.In the end,this paper studies the characteristics of two-phase high-resolution remote sensing images,and the two periods of data were superimposed and segmented.Through feature calculation and feature extraction,a feature image data with 96-dimensional features is constructed.By setting up multiple sets of comparison experiments,the effectiveness of the two proposed change detection algorithm frameworks are verified.Among them,the superposition algorithm based on SVDD feature selection and random forest has a simple structure.Compared with the traditional change detection algorithm,the accuracy is partially improved.Although the fusion algorithm based on SVDD feature selection and random forest is relatively complex,its accuracy is the best.