Change Detection And Tendency Analysis For Multi-temporal Remote Sensing Image

Change detection and tendency analysis for multi-temporal remote sensing images are important content for multi-temporal data interpretation.It contains many issues,such as where changed between multi-temporal data,how it changed,how many pixels changed and what the tendency is based on the aforementioned elements.With the development of the availability for multi-temporal data,the demand for change detection and tendency analysis have increased for many researchers.Firstly,they want to obtain the change result with high accuracy,meanwhile,reduce the pre-processing error for change detection.Secondly,they hope to understand the process of change detection so that they can acquire the attribute information for change regions.Finally,they expect to make full use of multi-temporal data and integrate the change result to achieve the tendency analysis and obtain the unknown temporal information.Therefore,the research on change detection for multi-temporal images conducted in this thesis is as follows.Firstly,the basic principles for multi-temporal images registration and its importance for the change detection pre-processing are introduced.In order to obtain sub-pixel level accuracy and decrease the influence of registration error for the multitemporal images change detection,the images from the same senor using the modified SIFT algorithm and the images from the different sensors adopting the histogram of orientated phase congruency algorithm which consider the structural properties are researched respectively.The accuracy and interference problems for different spatial resolutions multitemporal images change detection often exist obviously.Based on the introduction of the basic principle of multi-temporal image change detection,especially in the study of the traditional detection algorithm,the collaborative saliency model and color space smoothing algorithm are introduced in this thesis.The experiment results reveal that it not only obtains change results with high accuracy,but also overcomes the influence of noise,shadow,and error accumulation on the results.Meanwhile,it provides guarantee for subsequent change analysis.Change result only provide the location where have been changed between multi-temporal images.But in some cases,it is necessary to understand the properties of change regions and predict the distribution of the properties within the unknown intervals.Hence,for further analysis of changes,identifying the change classes and obtaining the distribution of the classes in the unknown temporal image,we have combined detection results and properties,meanwhile,proposed a joint prediction model based on spatio-temporal information,which realizes the combined utilization of multi-temporal images and change detection results,solves the problem of how to be changed and tendency analysis in the process of change detection and further obtains rich change information.