Some Key Techniques On Statistic Processing And Analysis Of High Resolution Remote Sensing Image

With the rapid development of earth observation techniques,the world will come to the epoch of high resolution remote sensing.By now,the resolution of optical remote sensing images has reached sub-meter level.Also the space borne SAR images are already better than one meter level.Meanwhile,the re-visit period has been significantly shortened.A revolution of our daily life is taking place because of the high resolution remote sensing.On one hand,due to the improvement of space resolution of remote sensing images,the mass of the remote sensing data increases with an explosion.It is of great important to process the remote sensing images more efficiently and to enhance the reliability of information abstraction.On the other hand,although some relative successful methods have been developed for the processing of traditional optical remote sensing images with mid-resolution,the applicability of them are often not satisfied for images with sub-meter resolution.Particularly,most of these methods are not applicable for the high resolution SAR images at all,which in turn reduces the application value of high resolution remote sensing data.Thus,more attentions should be paid to the development of high resolution optical and SAR remote sensing images processing,which is one of the key solutions to break the so called "many but few" dilemma.In this thesis,we have made a deep study on some key techniques as for the processing and analysis of high resolution SAR and optical remote sensing images,and some new methods have been proposed respectively for the despeckling,segmentation and target detection of SAR images,as well as the dehazing of optical remote sensing images.The major contributions of this paper are:Based on the generalized guided filter,we propose a despeckling-method for the high resolution SAR images.Such generalized guided filter is deduced from the standard guided filter by substituting the linear weight kernel with the nonlinear one.In addition,we construct the method for the estimation of such nonlinear weight kernel and so called guided image respectively under the framework of the Bayesian non local means and the Maximum likelihood rule.Experimental results validate the effectiveness of our method,which outperforms several classical and state of the art methods with better speckle noised restraining and details preservation.We propose a method based on well initialized Chan-Vese model for the segmentation for high resolution SAR images.The Gabor filter banks are firstly introduced to enhance the original image,and then the GMM model is applied to describe the statistics of the enhanced image,following which a pre-segmentation is carried out with EM algorithm to initialize the Chan-Vese model.Experimental results with high resolution SAR images containing vehicles,ships and man-made constructions indicate that as for the segmentation accuracy and implementation efficiency,our method performs better than the MRF based method and multi-scale level set based method.We propose an automatic target detection method for high resolution SAR image based on the saliency with context-aware.Firstly,the context-aware procedure is constructed with the statistical analysis of the characteristics of SAR images,then under the framework of Gaussian pyramid decomposition,we present the method for the estimation of target saliency.Finally,the saliency image is segmented with an adaptive threshold to separate the target region from the background region.Experimental results with MSTAR dataset and TerraSAR-X imagery validate the effectiveness of our method for the saliency detection,outperforming the GVBS,FT and other state of the art methods with the recall and precision measures.In addition,our method also performs better than some universal methods for target detection.For the dehazing of high resolution optical remote sensing images,we propose a novel approach based on local properties analysis.The degradation model difference between remote sensing images and natural images is presented,which proves the reasonability about the application of linear model to dehaze the high resolution remote sensing image.Then three main local properties are analyzed for the estimation of the parameters of linear model for dehazing.Experimental results indicated that our method performs well for high resolution optical remotes sensing images,which contains different scenes and haze of various thickness.As for the performance about haze removal and detail preservation,our method works better than several traditional and state of the art methods.