Research On High Resolution Remote Sensing Image Change Detection Method Based On Segmentation And Multi-features

In recent years,with the development and improvement of science and technology,cities size and natural environment are constantly updating and changing,and the utilization rate of land resources keeps improving.The timely and accurate detection and monitoring of changes in ground features using remote sensing images have become one of the indispensable important means of environmental resource investigation and management,urban planning and development.With the improvement of remote sensing image resolution,although high-resolution remote sensing images can bring more details of ground objects,it also leads to similarities among different ground objects,which cannot be effectively distinguished by traditional detection methods and single features.Aiming at the above problems,Firstly,this thesis introduces the shortcomings of traditional segmentation and change detection methods.Secondly,under the method of object-oriented analysis,the initial segmentation based on superpixels and the region is merged after the combination optimization of several characteristics of the initial segmentation region.Finally,a decision-level fusion method is proposed to realize the change detection of remote sensing image based on the combination of object and pixel.The main content of this thesis are as the following:1.A region merging algorithm based on superpixel segmentation is proposed.First,texture features are added on the basis of the original simple linear iterative clustering algorithm(SLIC)to obtain the initial segmentation object of the image.Then extract the spectrum and texture in the object to calculate the similarity measure between the objects and build the nearest neighbor model.Finally,the merging cost function is calculated to merge the superpixels according to the graph model and combining the spectral and shape characteristics.Experimental results show that this method can flexibly adjust the segmentation scale,retain object details,and improve segmentation accuracy.2.A decision level fusion algorithm is proposed to solve the problem that the change detection effect is affected by the segmentation scale.Firstly,images in different period are segmented to obtain overlapping segmentation labels,Secondly,the feature selection method is used to screen the feature extracted from the two-phase image to get the optimal feature subset.Secondly,pixel-level and object-level detection results are obtained by direct comparison and classification comparison.Finally,the two detectionresults are combined to identify the change region and obtain the final change detection result.Experiments show that this method can obtain more accurate detection results and detect subtle changes in the scene with many features and irregular distribution.