Research On Object-oriented Change Detection Of High Resolution Remote Sensing Images

Change detection of remote sensing imagery means by the methods of processing and analyzing remote sensing imagery photographed in different times,get the changing information of objects on the images.The processing objects of traditional change detection methods is mainly pixels on the imagery.The detection results obtained in the high-resolution remote sensing images are often not accurate.The object-oriented processing and analysis technology has become one of the internal components of the research on the change detection of remote sensing imagery in recent years.The object-oriented change detection technology of remote sensing imagery on is systematically studied.Image segmentation and multi-feature extraction were studied.A method of change detection of high-resolution remote sensing image based on multi-scale segmentation PCA and CVA was proposed.The major innovations of this dissertation are listed as follows:1.The research status,general principles,methods,classifications and main processes of remote sensing image change detection was studied and summarized.The characteristics of the current high resolution remote sensing image were summarized.The difficulties in the traditional change detection were analyzed.The research status,the principle,the development status and the general process of the object-oriented change detection of remote sensing images was studied.2.The principles,methods,classifications and main processes of image segmentation for remote sensing imagery were summarized.The segmentation of high-resolution remote sensing images based on Mean-Shift algorithm was realized.According to the concept and principles of the scale,the influence of scale on change detection was studied.Not every scale segmentation can improve the accuracy of change detection.The different processes and different results between multileveled segmentation and multiple scale segmentation were clearly pointed out.There was a hierarchical network structure in the results of multileveled segmentation,and the larger scale segmentation is based on the results of smaller scale segmentation.For the processes of multiple scale segmentation are independent,the results of multiple scale segmentation are independent.There is no clear relationship of hierarchical network in the segmentation results.3.The mean value,gradient intensity,shape index and LBP texture of the image object was selected as spectral features,spatial features,shape features and texture features,which reflects the information of remote sensing images.In order to study the Influence of these four features on change detection,a group of panchromatic images was detected by the subtractive method or CVA method based on the combination of the features in different types and different numbers,and the results was evaluated and compared.The number of features is not positively correlated with the accuracy of change detection.The increase of the number of features does not mean the improvement of accuracy.It is more reliable to select the features which are suitable for the target and change detection.4.A method of change detection of high-resolution remote sensing image based on multi-scale segmentation PCA and CVA was proposed.Based on image multi-scale segmentation and extraction of mean value of image objects,get the first N principal components by PCA transform.The principal components are calculated by CVA method to get the difference image.The difference image was segmented using Otsu method to get the change result image.The applicability of the method in change detection of panchromatic remote sensing images was compared by comparison experiments.It can effectively reduce the number and probability of pixels in false detection including commission and omission,while sustaining the probability of quality,so as to improve the accuracy and reliability of change detection.In addition,the change detection method based on multileveled segmentation is more dependent on the results of the first scale segmentation.While in the change detection method based on multiple scale segmentation,segmentation is carried out independently and the change detection results are more stable.