Unsupervised Polarimetric SAR Terrain Classification Based On Hierarchical Semantic Model And Scattering Mechanism

Polarimetric synthetic aperture radar(PolSAR)image classification is the main task for Pol-SAR image processing,and it is the premise of the PolSAR image understanding and inter-pretation.Compared with SAR images,PolSAR images contain more terrain information,and it is obtained by a multi-parameter and multi-channel imaging radar system.However,for heterogeneous regions,it is still a challenge to classify the aggregated terrain type,such as the urban area and the forests,into a semantic homogenous region due to their complex data and different structures.The aggregated terrain types are formulated by similar terrain objects aggregated together.These regions will formulate sharp bright-dark variations due to the scattering differences between the object and the ground.It is a challenge to both classify the aggregated regions into semantic homogeneous regions and preserve the edge details for PolSAR image.Recently,conventional PolSAR classification methods are miain-ly focus on the scattering characteristics and the low-level image features.In order to further understand the PolSAR images,high-level semantic information should be exploited,and it is the key for image understanding and interpretation.In this dissertation,the scattering characteristics of PolSAR data are fully exploited,and a hierarchical semantic model is constructed from the view of vision.Furthermore,a series of effective PolSAR image classification methods are proposed and introduced as follows.1.LA polarimetric hierarchical semantic model is proposed.For polarimetric SAR(PolSAR)image classification,it is a challenge to classify the aggregated terrain types,such as the ur-ban area,into semantic homogenous regions due to sharp bright-dark variations in intensity.The aggregated terrain type is formulated by the similar ground objects aggregated together.In this paper,a polarimetric hierarchical semantic model(HSM)is proposed to overcome this disadvantage by constructing a primal-level and a middle-level semantic.Firstly,the primal-level semantic is a polarimetric sketch map which consists of sketch lines as the s-parse representation of a PolSAR image.Secondly,the middle-level semantic is a region map which is proposed to extract semantic homogenous regions from the sketch map.It is obtained by exploiting the local spatial relationship between each sketch segment and its neighbor segments.The region map can partition a PolSAR image into aggregated,structural and homogenous region types.The experimental results indicate that the proposed semantic model can partition a PolSAR image into three region types with different structures well.2.A new hierarchical semantic model and polarimetirc characteristics based PolSAR image classification method is proposed.In order to classify the complex PolS AR scene accurately,we propose a new hierarchical semantic model and polarimetric characteristics based Pol-SAR image classification method.Mapping the region map to the PolSAR image,a com-plex PolSAR scene is partitioned into aggregated,structural and homogenous pixel-level subspaces with the characteristics of relati,vely consistent terrain types in each subspace.Initial segmentation is obtained by the mean shift(MS)method,and then three subspaces are merged by different merging schemes respectively,and further a polarimetric-semantic clas-sifier is designed to improve the classification result.Some experiments are taken on four data sets with different bands and sensors,and the experimental results indicate that the pro-posed method can obtain semantic homogeneous regions and edge details for polarimetric SAR image classification.3.An unsupervised polarimetric synthetic aperture radar image classification method based on sketch map and adaptive markov random field is proposed.Markov random field(MRF)model is an effective tool for polarimetric synthetic aperture radar(PolSAR)image classi-fication.However,due to the lack of suitable contextual information in conventional MRF methods,there is usually a contradiction between edge preservation and region homogene-ity in the classification result.To preserve edge details and obtain homogeneous regions simultaneously,an adaptive MRF framework is proposed based on a polarimetric sketch map.The polarimetric sketch map can provide the edge positions and edge directions in detail,which can guide the selection of neighborhood structures.Specifically,the polari-metric sketch map is extracted to partition a PolSAR image into structural and nonstructural parts,and then adaptive neighborhoods are learned for two parts.For structural areas,ge-ometric weighted neighborhood structures are constructed to preserve image details.For nonstructural areas,the maximum homogeneous regions are obtained to improve the region homogeneity.Experiments are taken on both the simulated and real PolSAR data,and the experimental results illustrate that the proposed method can obtain better performance on both region homogeneity and edge preservation than the state-of-the-art methods.4.A deep learning and hierarchical semantic model based PolSAR terrain classification method is proposed.Stacked auto-encoder model can effectively represent the complex terrain structures,such as the urban and the forest,by automatically learning the high-level features.However,it has difficulty in preserving details and edges.In order to overcome this shortcoming,a new unsupervised polarimetric SAR classification method is proposed by combining the deep learning and the hierarchical semantic model.According to the HSM,a PolSAR image is partitioned into aggregated,homogeneous and structural regions.For aggregated regions,a stacked auto-encoder model is applied to learn high-level features,and further the sparse representation and classification is constructed by learning a dictio-nary with high-level features.For homogeneous regions,a hierarchical segmentation and classification is applied.In addition,edges are located and line objects are preserved for structural regions.Experimental results demonstrate that the proposed method can obtain good performance in both region homogeneity and edge preservation.5.A novel edge detection method for PolSAR images based on wavelet fusion is proposed.Polarimetric constant false alarm rate(CFAR)edge detector can suppress the speckle noises by considering the Wishart distribution of Polarimetric SAR data.However,it has difficulty in detecting the edge details in the heterogeneous regions such as the thin roads in urban area since the homogeneity assumption in a filter is not satisfied any more.In order to overcome this disadvantage,a novel edge detector is proposed by combining the merits of the polarimetric CFAR detector and weighted gradient-based detector which can detect details well.Wavelet transformation is utilized and semantic rules are defined to fuse the two complementary detectors.Moreover,a despeckling scheme is applied to reduce the speckle noise from the gradient-based detector.Experimental results demonstrate that the proposed method can obtain sound performance in both weak edges and heterogeneous regions.