Research On Change Detection Method For High Resolution Remote Sensing Images

Change detection is used to extract the difference characteristics between different periods,analyze the feature differences,and finally detect and identify the change regions(change information)between different periods by comparing and analyzing the images of the same position captured in different periods.The research object of this thesis is mainly high-resolution remote sensing images,specifically how to analyze quantitatively and determine the surface changes from remote sensing data of different periods.With the development of space technology,high-resolution images are widely used,because they contain more abundant ground object information and have the advantages of high spatial resolution,high definition and so on.In recent years,change detection approaches of remote sensing image has been widely used in land cover and utilization,natural disaster prediction and assessment,agricultural resources investigation,urban planning and layout,resource investigation,environmental monitoring and analysis,urban expansion and change information acquisition,geographic data update,military reconnaissance and other fields.Therefore,it is of great application value to study the research on change detection methods for high-resolution remote sensing images.The main work and contents of this thesis are as follows:(1)In this thesis,the author deeply investigates and studies the references of achieving change detection of recent years,then analyzes the mainstream change detection methods.Combined with the characteristics of high-resolution remote sensing images,it proves that it is necessary to study the change detection methods which is suitable for high-resolution remote sensing images.After that,it has found out the problems existing in change detection through investigating and analyzing the research,and then propose an executable solution for the specific problem,so as to better realize the change detection in future.(2)For the traditional change detection methods,the first problem is that threshold image segmentation is used to obtain the change region based on the gray-scale difference image,and the structural information problem of the high-resolution remote sensing image is neglected;the second problem is that the high-resolution remote sensing image size is too large,which leads to a higher execution time.In this thesis,an unsupervised change detection method based on fast fuzzy C-means clustering is proposed.The improved fuzzy C-means clustering algorithm is applied to very high-resolution(VHR)remote sensing images,which can effectively utilize the spatial information of images.Further obtaining a better difference image that has more representative of the difference feature;On the other hand,the method combines Gaussian pyramid and fuzzy C-means clustering algorithm,which aims to effectively reduce the redundant data of the image,improve the computational efficiency of the algorithm,and reduce the computational complexity.(3)In this thesis,it also proposes a change detection framework based on deep convolutional neural network,which is used to obtain the changed regions of high-resolution remote sensing images.The model is a U-shaped symmetrical structure based on U-net network.It can use multiple layers of connections,consider features from multiple convolutional layers,merge low-dimensional and high-dimensional features into the final feature map,and effectively explore the image.It helps to gain a domain with wider acceptance.Moreover,the multi-scale pooling module can effectively utilize the spatial multi-scale features of landslide areas,and effectively solve the shortcomings of global pooling.The proposed deep convolutional neural network model for change detection provides more accurate localization for change detection.