Study On Reliable Change Detection Methods For Very High Resolution Remote Sensing Images Based On Difference Measures

Remote sensing image change detection(RSICD)is a technique for detecting and analyzing the change information using remote sensing images in different periods.It has been widely used in the fields of agricultural investigation,forest monitoring,disaster assessment,urban expansion,and environmental monitoring.This technique has become an effective means to study the change of land cover and is also one of the hotspots in the field of remote sensing.However,due to the complexity of the surface environment,there is a systematic deviation in the spectral reflection value of the object.Especially with the increasing of spatial resolution of remote sensing images,the spectral separability of different objects decreases gradually,which affects the reliability of change detection results.To address the above problem,this dissertation analyzes the reliability in the process of RSICD to improve the reliability of the change detection results.This dissertation mainly focuses on the stage of the difference image generation in RSICD and puts forward some difference measures.On the basis,new reliability change detection methods are constructed,and the difference image-based reliable change detection theories are further improved,which provides theoretical support for the application of change detection.The details are as follows:(1)The technical system and reliability of RSICD are systematically discussed in the dissertation,such as the reliability in data preprocessing,difference image generation,change information extraction and accuracy evaluation,etc.,the dissertation will put forward three strategies for generating the reliable difference image,namely,the spatial information-based difference measure,the spectral and texture informationfused difference measure and remote sensing images with different spatial resolutionbased difference measure.(2)On the basis of fully analyzing the shortcomings of the traditional spatial information-based change detection methods,such as the extraction of spatial information and the difference processing methods,a Gabor wavelet difference measure-based change detection method for very high resolution(VHR)remote sensing images is proposed.In the method,the difference image is generated by the similarity between two temporal Gabor wavelet feature images described by Markov random field model.The experimental results show that the proposed method is robust to noise and can achieve higher detection accuracy,and hence enhance the reliability of the change detection results.(3)In the traditional spectral and texture features-based RSICD,the fusion strategy for spectral and texture information and the change information extraction method are insufficient.In this dissertation,a new RSICD method,using spectral and gray level co-occurrence matrix(GLCM)texture features difference measure,is proposed.Based on the principle of similarity matching,the difference image is obtained by comparing the similarity of two temporal texture feature images in a certain space distance.On this basis,a new change detection process is designed to fully utilize the spectral and spatial information,in which the spectral feature image and GLCM texture feature images are processed separately.Finally,an improved active contour model based on expectation maximum method is proposed to extract the change information from the difference image.The experimental results show that this method can effectively improve the reliability of the change detection results.(4)In the subpixel mapping-based change detection methods for multisensor images with different spatial resolution,the result is always affected by the accuracy of mixed pixel decomposition.In order to solve this problem,a new change detection method for remote sensing images with different spatial resolution is proposed based on fraction image difference measure.In this method,the fraction image change detection and subpixel change detection are used to generate the final change detection results.In the fraction image change detection,the Gaussian mixture distribution model is used to process two temporal fraction images.And then,the spatial attraction model is utilized to allocate the subpixels and subpixel change detection.Finally,the change detection results are obtained.The experimental results show that the proposed method can effectively reduce the uncertainty in the process of mixed pixel decomposition,obtain higher detection accuracy,and hence improve the reliability of the change detection results.