Research On Some Key Technologies Of Remote Sensing Image Processing

Remote sensing image is the product of various sensors to obtain information, and provides spectralinformation and abundant spatial about the observed object. It has been widely used in civil andmilitary fields. Aiming at overcoming the shortages of the existing remote sensing image processingmethods, research on processing methods that are suitable for the characteristics of remote sensingimages has important theoretical meanings and application value. This paper studies remote sensingimage processing technology such as image matching, image fusion, edge detection, imageclassification, change detection. Main work is as follows:Firstly, a remote sensing image matching method based on contourlet transform and speed uprobust features (SURF) is discussed. The standard image and the registered image are decomposed bycontourlet transform. The decomposed low frequency images are inputted to SURF algorithm toobtain the coarse matching results. And the random sample consensus (RANSAC) algorithm is usedto wipe out the mismatching point pairs. The experimental results show that the method has fasteroperation speed and higher matching accuracy than SIFT and SURF.Then an infrared and visible image fusion method based on no nsubsampled shearlet transform(NSST) and weighted non-negative matrix factorization (WNMF) is proposed. Nonsubsampledshearlet transform is used to decompose infrared and visible images. Dynamic weighted non-negativematrix factorization algorithm is adopted for fusion processing of low-frequency coefficients. And thefusion rule based on maximum of absolute values is used for high-frequency coefficients. Theexperimental results show that the method has better results than others.And then, an edge detection method of SAR image based on anisotropic diffusion, nonsubsampledcontourlet transform(NSCT) and improved mathematical morphology is given. The image is firstlydecomposed by NSCT. Then different diffusion models are carried on the low frequency part and thehigh frequency part. The improved mathematical morphology method is used to process the low-frequency image, and the method of modulus maxima is applied to the high-frequency image to detectedges. The whole edges are obtained by fusion. The experiments show that this method has betteredge detection performance than Canny method, Sobel method, Prewitt method, the method ofwavelet modulus maxima.Next, a land cover remote sensing image classification method based on Krawtchouk momentinvariants, Log-gabor and SVM is presented. Firstly, Log-gabor filter is used to extract texture featureof remote sensing images. Then Krawtchouk moments are combined to construct edge shape characteristic. SVM is used to classify the extracted feature vector. Experimental results show that,compared with the Gabor method, Log-gabor method, krawtchouk moments method, the proposedmethod has higher accuracy.Finally, a change detection method based on wavelet transform and kernel independent componentanalysis (KICA) is proposed. Remote sensing images are decomposed by wavelet transform, and thedecomposed data is analysed by KICA. According to the differences between the high-frequencycomponents of separated vectors, the change component is distinguished automatically. Experimentalresults show that, the proposed method can separate change information of remote sensing imageswith higher accuracy, and the intelligent change detection is realized.