Research Of Cloud Removing Method For Remote Sensing Image Based On Support Vector Machine

Support vector machine is a machine learning approach which has powerful ability to captive verge point. Support vector value contourlet transform, which is constructed by support vector machine, is a new kind of image representations which is of multi-scale, multi-direction and shift-invariance. It is implemented by two stages. Firstly multi-scale decomposition of the source image is performed with support vector value filter, which is given by regression model of support vector machine. Then directional decomposition of the high frequency subbands in every scale is made with nonsubsampled directional filter banks. When introduced into image processing, support vector value contourlet transform can better expressed detail feature of image with prejudice to keep geometric feature of image than other multi-scale transforms.Cloud removal is a vital task of image preprocessing. By studying the principle of support vector machine and its application on thin-cloud & thick-cloud, the main work and production of this paper are achieved as follows:1. The development of support vector machine and was the general condition of multi-resolution analytical approach were introduced. The theory, implementation of support vector value contourlet transform was particularly discussed.2. The theory, implementation of thin-cloud removal was studied. By comparing the main difference of information between the clouds and ground features, a method for thin-cloud removal based on support vector value contourlet transform was proposed. By using support vector value contourlet transform, cloud information was filtered to obtain the ground information. The ground information was enhanced with the multi-scale coefficients, thresholds and the enhancement functions which were adaptively set in the algorithm. Finally gray value distribution was adjusted by histogram matching to get the result. By making research on color model of image, the similar process was adopted in removing thin cloud of color image. Compared with nonsubsampled contourlet transform and homomorphic filter, the method can effectively keep the geometric feature of ground and eliminate the contamination of thin-cloud in the experiment.3. The theory, implementation of thick-cloud removal was studied. A method for thick-cloud removal based on support vector machine was proposed. The rule graph of cloud layer and shadow was generated by the detection of cloud layer and shadow using support vector machine and information of solar angle. Selection matrices are constructed by logical decision of the rule graph of cloud layer and shadow among multi temporal remote sensing images. Preliminary cloud layer and shadow removal was achieved by image mosaic based on support vector value contourlet transform using selection matrices. The overlap of cloud layer and shadow was repaired to remove by the method of high and low frequency compensating. Thick-cloud removal was accomplished by reconstructing image and using median filter. Compared with methods of wavelet transform, the method using in this paper can better reappear the ground information of cloud covering and have better image smoothness and image definition.