Sparse Representation And Manifold Learning Based Method For SAR Image Classification

Synthetic aperture radar(SAR)is an active imaging radar system operating in all-weather and all-time conditions,can obtain high quality remote sensing images from airborne or spaceborne platforms.This system is widely used in military and civil fields with the ability to detect targets.And the multichannel SAR is called Pol SAR.It alternately transmits and receives radar signal to obtain more complicated scattering information.Compared to the single channel SAR,Pol SAR can get stronger perception from land covers and more abundant feature information.This thesis aims at solving SAR land cover classification problems and developing sparse representation classifiers for complicated scenes.And it combines the sparse representation and manifold learning relevant algorithms to mine target information from polarimetric scattering data,polarimetric target decompositions,and polarimetric image.Based on above algorithms,we did some works about Pol SAR feature extraction and fusion,and unsupervised classification.(1)A novel hierarchical sparse representation-based classification(HSRC)for SAR images is proposed.Features utilized in HSRC are extracted from the multi-size patches around each pixel to precisely describe the complex terrains.Two thresholds are introduced in the sparse representation classifier to restrict the range of reconstruction residual,which classifies the reliable classified points,and the rest of the pixels are considered as the uncertain ones in the original SAR image.Then,a new dictionary is constructed by the reliable pixels,and the uncertain pixels will be reclassified in the next classification layer.The hierarchical structure is very reasonable and effective to employ simple features in each layer for describing the various topographic types.Compared with traditional sparse representation-based classification and support vector machines in several fixed-size patches,the proposed method can obtain better performance both in quantitative evaluation and visualization results.(2)A novel Pol SAR image unsupervised classification method is proposed.It combines three typical features,including polarimetric data features(coherent matrix),polarimetric decomposition features(Krogager,Freeman,Yamaguchi,Neumann,and H/A/ adecomposition),and gray-level co-occurrence matrix features to comprehensively describe the data characteristics.And it also proposes a symmetric revised Wishart(SRW) distance-derived manifold regularized low-rank representation(SRWM_LRR)method to deeply exploit the geometry data structure.The low-rank representation(LRR)is used to capture the intrinsic global structure of Pol SAR data and the manifold regularization is employed to detect the local structure of the data,in which SRW distance is introduced to measure the similarity between different pixels for describing the local manifold structure.This algorithm considers the specific statistics property in Pol SAR data and simultaneously integrates multiple features in perspective of data geometry structure to represent pixels for achieving a better classification performance.The effectiveness and practicability of the proposed method are demonstrated by datasets obtained either in spaceborne or airborne SAR system,including the Flevoland dataset(AIRSAR L-Band)extensively used in land cover classification,and Xi'an dataset(RADARSAT-2 C-Band).Compared with the traditional Wishart classifier,Euclidean and SRW distance-based spectral clustering and LRR,the proposed method shows an improvement in accuracy and efficiency as well as a better visualization result.(3)A novel Pol SAR feature fusion method is proposed.It is well known that various features extraction approaches are utilized in Pol SAR terrain classification for representing the data characteristic.It needs relevant and effective feature fusion algorithms to process complicated features.To address this issue,we present a multimodal sparse representation(MSR)framework based algorithm to fuse the different feature vectors from the complicated data space.Polarimetric data feature,decomposition feature,and the texture feature from Pauli color-image are selected to represent multimodal data in different observation modes.The corresponding multimodal manifold regularizations are added to MSR framework to approximate the data structure.Considering the independence and correlation of features,the calculated intrinsic affinity matrices are calculated from this framework.Then it is processed by the preserve projection(LPP)algorithm to project the multimodal features into a low dimensionally intrinsic feature space for subsequent classification.Effectiveness and validity of proposed method are demonstrated in three datasets from the RADARSAT-2 system in C-band: Western Xi'an,Flevoland and San Francisco Bay region.(4)We propose a novel subspace cluster based method to process the Pol SAR imagery classification.The reduced features and clustering effect are analyzed under different regularization constraints.First,the related comprehensive feature vectors are extracted from Pol SAR data.They cover many useful decomposition features including the power of components in coherent(Krogager)and incoherent decompositions(Freeman,Yamaguchi,Van Zyl,Neumann),and the parameters with definitely physical characteristics from H/A/a and Touzi decompositions.Most of them can be easily realized by Pol SARpro5.0 software.Second,the projection matrix and affinity matrix can be calculated simultaneously under the constraints of sparse representation,LLR,and manifold regularization terms.The sparse representation corresponds to select a few of points form the same subspace,LLR to capture the global structure of samples,and regularization terms to detect the local manifold structure of data.The algorithm aims at finding groups of data points from different subspace in samples space.Experiments demonstrated that the subspace clustering method is valid to Pol SAR land cover classification.