Hierarchical Visual Computation And Statistical Model Based SAR Image Segmentation And Understanding

SAR is an active imaging system with microwave which can be operated day and night under all weather conditions.It has become the main monitoring means of the ground activities.At the present,massive quantities of SAR images are available due to the rapid development of the SAR imaging technique.However,the capacity of the SAR image processing is not experiencing a synchronous growth.Therefore,extracting useful information from the massive SAR images is an urgent task.It means that SAR image interpretation is necessary.SAR image segmentation and understanding are critical step in the SAR image interpretation.Hence,it is of great significance to promote the development of SAR image processing technology by SAR image segmentation and understanding.In the thesis,we propose a new framework for SAR image processing based on hierarchical visual computation and statistical model.The new framework combines the semantic space derived from the Marr's computer vision and the pixel space represented by the statistical model.In the framework,our group proposes hierarchical visual computation model based on the Marr's computer vision.Under the guidance of the semantic space,we plug the semantic information to the statistical model for the segmentation of the SAR image.With the prior information and the domain knowledge of the SAR image,a new SAR image understanding method is constructed through the alternant work of the semantic space and the pixel space.Under the new framework,we have five works.They are described as follows.(1)MRF model has been successfully applied to SAR image segmentation because of its excellent ability of capturing the local contextual information in the prior model.However,the geometric structures of the SAR image are always ignored when capturing the contextual information in the prior model.Therefore,we present a new SAR image segmentation method based on sketching model and higher-order neighborhood Markov random field.In this approach,the sketching model is utilized to represent the geometric structures of the SAR image.Meanwhile,a higher-order neighborhood is constructed to capture the complex priors.Then,according to the structure fluctuation in the higher-order neighborhood,the homogeneous and heterogeneous neighborhoods are distinguished.Finally,the local energy function in the prior model is constructed in the higher-order neighborhood with different characteristics.Specifically,the energy functions considering the labeling consistency and focusing on the structure preservations are designed for the homogeneous and heterogeneous neighborhoods,respectively.In this way,the ability of the prior model is improved by adding the geometric structures into the energy functions.Experiments on the real SAR images demonstrate the effectiveness of the proposed method in labeling consistency and structure preservations.(2)Recently,the multinomial latent model has been successfully applied to SAR image segmentation.However,the single spatial relationship is difficult to deal with the heterogeneous structures of the SAR images.We propose an adaptive hierarchical multinomial latent model with hybrid kernel function for SAR image segmentation.In the proposed scheme,we design a hybrid kernel function combing Gaussian radial basis function(GRBF)and ridgelet kernel function to adaptively describe the spatial relationship between the central pixel and the surrounding pixels.Specifically,the SAR image is divided into nonstructural region and structural region by SAR sketching model.For the non-structural region,in order to improve the labeling consistency and reduce the wrong segmentation,we construct the multi-layer multinomial latent model with GRBF for segmentation.For the structural region,in order to preserve the details(such as edge,lines,small objects),we adopt the single-layer multinomial latent model with ridgelet kernel function for segmentation.Finally,the segmentation results of non-structural and structural regions are integrated together to obtain the final segmentation result.Comprehensive experiments on both synthetic and real SAR images demonstrate that the segmentation results by our proposed scheme achieve the labeling consistency and details preservations simultaneously.(3)SAR imaging system usually produces pairs of bright area and dark area when depicting the ground objects,such as,a building or tree and its shadow.Many buildings(trees)are aggregated together to form urban areas(forests).It means that the pairs of bright and dark areas often exist in the aggregated scenes.Conventional unsupervised segmentation approaches usually segment the scenes(e.g.,urban areas and forests)into different regions simply according to the gray values of the image.However,a more convincing way is to regard them as the consistent regions.We aim at addressing this issue and propose a new SAR image segmentation approach via hierarchical visual semantic and adaptive neighborhood multinomial latent model.In this approach,hierarchical visual semantic of SAR images is proposed,which divides SAR images into aggregated,structural and homogeneous regions.Based on the division,different segmentation methods are chosen for these regions with different characteristics.Moreover,we design a visual semantic rule to locate the line objects in the SAR image.Finally,these results are integrated together to obtain the final segmentation.Experiments on both synthetic and real SAR images indicate that the proposed method achieves promising performances in terms of the consistencies of the regions and the preservations of the edges and line objects.(4)SAR imaging system is usually an observation of the earths' surface.It means that rich structures exist in SAR images.Convolutional neural network(CNN)is good at learning features from raw data automatically,especially the structural features.Inspired by these,we propose a novel SAR image segmentation method based on convolutional-wavelet neural networks(CWNN)and MRF.In this approach,a wavelet constrained pooling layer is designed to replace the conventional pooling in CNN.The new architecture can suppress the noise and is better at keeping the structures of the learned features,which are crucial to the segmentation tasks.CWNN produces the segmentation map by patch-by-patch scanning.The segmentation result of CWNN will be used with two labeling strategies(i.e.,a superpixel approach and a MRF approach)to produce the final segmentation map.Experiments on the texture images demonstrate that the CWNN is effective for the segmentation tasks.Moreover,the experiments on the real SAR images show that our approach obtains the regions with labeling consistency and preserves the edges and details at the same time.(5)SAR image understanding is a high level description for SAR scenes.It does not require just dividing the SAR image into individual regions,but it also helps in describing the terrain types of the regions.For example,the terrain type of one region is the water,and the terrain type of another region is the urban area,and so on.In order to understand the SAR image accurately,we propose a SAR image understanding approach based on semantic conditional random fields and Bayesian networks.In the proposed approach,a SAR image understanding framework is proposed.In the new framework,the semantic space interacts with the pixel space in order to ensure obtaining an accurate SAR image understanding.Firstly,the SAR image is divided into aggregated,structural and homogeneous regions which are implemented by the regional map in the semantic space.The aggregated region is segmented by using the bag of words method.The structural and homogeneous regions are segmented by plugging the semantic information into conditional random fields(CRFs).Secondly,with the semantic segmentation results,a Bayesian network(BN)is constructed in order to model the statistical dependencies that have arisen among the regions and their corresponding terrain types.According to the prior knowledge and the domain knowledge,BN is used to infer the terrain types of most regions.Finally,the SAR image understanding results are fed back to the semantic space in order to generate a higher-level semantic,which is useful for the higher-level understanding of the SAR image.Experiments on the real SAR images demonstrate and emphasize that the proposed semantic segmentation method produces the semantic segmentation result,which achieve the good labeling consistency and accurate detail preservations simultaneously.Moreover,the proposed SAR image understanding method gives the SAR image understanding result and gives the terrain types of the most regions in a probability way.