Research On Some Key Technologies Of High Resolution Remote Sensing Image Processing

High resolution remote sensing image provides abundant spatial and spectral information about theobserved object, and it has been widely used in military and civil fields. In order to overcome theshortcomings of the existing remote sensing image processing methods, there are importanttheoretical meanings and application value to develop more efficient analysis and processing methodsthat are suitable for the characteristics of high resolution remote sensing image. On the basis ofprevious research results, researches on high resolution remote sensing image segmentation, match,fusion, change detection and road detection have been done in this paper, and described as follows:Firstly, we discuss a kernel fuzzy C-means segmentation method based on local spatial information.Not only the use of the kernel function clustering of data to be mapped to the high-dimensional space,integrated pixel spatial information. The method can be directly applied to the original remote sensingimage, and does not require filtering pretreatment. The experimental results show that the method hasbetter homogeneous regions than FCM, FLICM and KFCM.Then a remote sensing image matching method based on nonsubsampled contourlet transform(NSCT) and speed up robust features(SURF) is given. Two related low frequency images decomposedwith NSCT are inputted to SURF algorithm to obtain the pre match result. And the missmatching iswiped out by using the random sample consensus(RANSAC) algorithm. This method increases thematching speed as well as improves the accuracy of the match.And then, a fusion method for remote sensing images based on improved gradient projectionnon-negative matrix factorization(NMF) and complex contourlet transform(CCT) is studied. In thismethod, the CCT coefficients of high and low frequency subbands are fused by different rules. Andimproved gradient projection optimization method can reduce the NMF iteration time complexity.Evaluation based on both subjective and objective criteria shows that the proposed method hassuperior results than others.Next, multitemporal remote sensing images change detection based on kernel principal componentanalysis(KPCA) and kernel independent component analysis(KICA) are introduced. KPCA/KICAextracting principal/independent component with nonlinear method can be improved by PCA/ICA.They are suitable for the image characteristics of remote sensing images. A large number ofexperiments show that this method can effectively detect the change regions in the high resolutionremote sensing images.Finally, a road detection method from urban remote sensing image based on KFLICM and mathematical morphology is presented. A component image containing most of the road informationis segmented from original image by KFLICM. Then use a series of mathematical morphologyprocessing to obtain the road target. Experimental results show that most of urban main roads can beaccurately detected by proposed method.