Search

Removing redundant features and improving classifier performance necessitates the use of meta-heuristic and deep learning (DL) algorithms in feature selection and classification problems. With the maturity of DL tools, many data-driven polarimetric synthetic aperture radar (POLSAR) representation models have been suggested, most of which are based on deep convolutional neural networks (DCNNs). In this paper, we propose a hybrid approach of a new multi-objective binary chimp optimization algorithm (MOBChOA) and DCNN for optimal feature selection. We implemented the proposed method to classify POLSAR images from San Francisco, USA. To do so, we first performed the necessary preprocessing, including speckle reduction, radiometric calibration, and feature extraction. After that, we implemented the proposed MOBChOA for optimal feature selection. Finally, we trained the fully connected DCNN to classify the pixels into specific land-cover labels. We evaluated the performance of the proposed MOBChOA-DCNN in comparison with nine competitive methods. Our experimental results with the POLSAR image datasets show that the proposed architecture had a great performance for different important optimization parameters. The proposed MOBChOA-DCNN provided fewer features (27) and the highest overall accuracy. The overall accuracy values of MOBChOA-DCNN on the training and validation datasets were 96.89% and 96.13%, respectively, which were the best results. The overall accuracy of SVM was 89.30%, which was the worst result. The results of the proposed MOBChOA on two real-world benchmark problems were also better than the results with the other methods. Furthermore, it was shown that the MOBChOA-DCNN performed better than methods from previous studies. [ABSTRACT FROM AUTHOR]

Highly accurate supervised deep learning-based classifiers for polarimetric synthetic aperture radar (PolSAR) images require large amounts of data with manual annotations. Unfortunately, the complex echo imaging mechanism results in a high labeling cost for PolSAR images. Extracting and transferring knowledge to utilize the existing labeled data to the fullest extent is a viable approach in such circumstances. To this end, we are introducing unsupervised deep adversarial domain adaptation (ADA) into PolSAR image classification for the first time. In contrast to the standard learning paradigm, in this study, the deep learning model is trained on labeled data from a source domain and unlabeled data from a related but distinct target domain. The purpose of this is to extract domain-invariant features and generalize them to the target domain. Although the feature transferability of ADA methods can be ensured through adversarial training to align the feature distributions of source and target domains, improving feature discriminability remains a crucial issue. In this paper, we propose a novel polarimetric scattering characteristics-guided adversarial network (PSCAN) for unsupervised PolSAR image classification. Compared with classical ADA methods, we designed an auxiliary task for PSCAN based on the polarimetric scattering characteristics-guided pseudo-label construction. This approach utilizes the rich information contained in the PolSAR data itself, without the need for expensive manual annotations or complex automatic labeling mechanisms. During the training of PSCAN, the auxiliary task receives category semantic information from pseudo-labels and helps promote the discriminability of the learned domain-invariant features, thereby enabling the model to have a better target prediction function. The effectiveness of the proposed method was demonstrated using data captured with different PolSAR systems in the San Francisco and Qingdao areas. Experimental results show that the proposed method can obtain satisfactory unsupervised classification results. [ABSTRACT FROM AUTHOR]

As a special dual-polarization technique, compact polarimetric (CP) synthetic aperture radar (SAR) has already been widely studied and installed on some spaceborne systems due to its superiority to quad-polarization; moreover, quad-pol information can be explored and reconstructed from the CP SAR data. In this paper, a refined model is proposed to estimate the quad-pol information for the CP mode. This model involves CP decomposition, wherein the polarization degree is introduced as the volume scattering model parameter. Moreover, a power-weighted model for the co-polarized coherence coefficient is proposed to avoid the iterative approach in pseudo-quad-pol information reconstruction. Experiments were implemented on the simulated Gaofen-3 and ALOS-2 data collected over San Francisco. Compared with typical reconstruction models, the proposed refined model shows its superiority in estimating the quad-pol information. Furthermore, terrain classification experiments using a complex-value convolutional neural network (CV-CNN) were performed on AIRSAR Flevoland data to validate the reconstruction effectiveness for classification applications. [ABSTRACT FROM AUTHOR]

Polarimetric Synthetic Aperture Radar (PolSAR) is one of the most important remote sensing tools. However, PolSAR images are strongly contaminated by a multidimensional interference (called speckle noise), making their processing (e.g. in the classification context) difficult. In terms of structure, multilook PolSAR data follow a definite positive hermitian behaviour and, therefore, require tailored classifiers for their features. Some classic classifiers – such as Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), K-Nearest Neighbours (KNN), and Support Vector Machine (SVM) – have been yielding unacceptable performance to these data, when applied directly. One justification is because they do not often take into account neither the speckle presence nor properties which are inherent to under-study relieves. This paper addresses a collection of PolSAR divergence-based classifiers, deduced from the normal, skew-normal, t-Student, and skew-t vector models as well as the scaled complex Wishart (SCW) distribution. The last model is a standard supposition to describe multilook PolSAR data, having two parameters: covariance matrix (which is directed to data nature) and number of looks (which controls the speckle noise effect). The considered remainder laws aim to model the main diagonal of these data, known as multivariate intensities. The performance of proposed methods is quantified and compared with those due to the Kullback-Leibler (KL) distance for multivariate normal distribution and to LDA, QDA, KNN, and SVM methods. Experiments with both artificial and real PolSAR data are considered. Results favour optimal Rényi classifiers for an Airborne Synthetic Aperture Radar (AIRSAR) image of San Francisco and the t-Student KL classifier for an SAR image system of the Electromagnetics Institute (EMISAR) image of Foulum. [ABSTRACT FROM AUTHOR]

For a long time, Polarimetric Synthetic Aperture Radar (PolSAR) data was not available free of cost, so the applications were limited. With the recent increase in availability of PolSAR data due to missions like ALOS-2, UAVSAR and Sentinel, the data can be acquired through these sensors periodically. Since the volume of data is large, applying traditional classifiers and conventional machine learning algorithms becomes confounding, as in that case, manual feature extraction must be done by the researcher for training the model. To automate this feature extraction step and accelerate the process, many researchers have used various deep learning algorithms. However, most studies fail to tap into the potential of PolSAR data by not utilizing the complete range of complex data of float-32 bit-depth. The work in this paper suggests the development of a generalized deep learning model for nine elements of coherency matrix [ T 3 ] in the float-32 data-space from ALOS/PALSAR-2 L-Band sensor. Semantic segmentation is used to classify land-covers into various classes like water, settlement, forest, open land, wetlands etc. This is done through modification on the UNet backbone. The deep learning model is trained using data on San Francisco and New Delhi regions and tested on data from Mumbai region. It is observed that the classification accuracy for Mumbai region is " 93.82 % ". This kind of system can assist the decision makers like urban planners to take informed decisions. [ABSTRACT FROM AUTHOR]

The assumption of reflection symmetry is one of the major deficiencies for model-based decompositions. The introduction of helix scattering components can mitigate some of the impact of this assumption while simultaneously generating some new negative power problems. The helix mechanism has been developed in the Krogager coherent target decomposition where there is no negative power problem. Why do these problems happen in model-based decompositions? In this paper, we review the Krogager decomposition based on the coherency matrix form of the unified CTD model and come to the conclusion that the imaginary part of T23 is an overestimate of the helix contribution, which is the main cause of the problem. Furthermore, we review the techniques of symmetric scatterer transformation in the Cameron and Huynen decompositions and expand them to the multilook coherency matrix. Two new models of the nonlinear programming problem are proposed to get the transformed coherency matrix: 1) by subtracting the asymmetric component; and 2) by the procedure of con-diagonalization. A new parameter is established to measure the degree of reflection asymmetry of the coherency matrices. Experimental results in the L-band Airborne SAR (AIRSAR) San Francisco data show that the two new models can reduce the correlations between copolarized and cross-polarized channels significantly. In this case, the assumption of reflection symmetry is reasonable for the transformed coherency matrices, and negative power problems related to helix scattering no longer exist in the model-based decompositions; hence, the helix scattering component is not needed. Lastly, the Freeman and Durden decomposition has been applied to the transformed coherency matrices. Experimental results showed that the decomposition results can reflect the predominant characteristics of the ground objects. [ABSTRACT FROM PUBLISHER]

Synthetic Aperture Radar (SAR) polarimetric datasets are widely used in the detection and classification of urban areas. Most methods used today are based on the decomposition of fully polarimetric SAR data, which allows for the extraction of physical information about the nature of the medium and the application of proper classification methods. According to the theory, the main and predominant backscattering mechanism for buildings is double bounce. However, when analyzing urban environments, the observed predominant backscatter may differ from theory depending on many aspects. In this paper, we analyze fully polarimetric ALOS PALSAR data for various cities located on different continents, proving that the theory does not hold for most cases. There are many factors that have an impact on the detected backscatter mechanism, and the theoretical principle of predominant double bounce in urban areas can be met only under specific conditions. These factors are, among others, the orientation of the buildings, the dimensions of the streets, the type of construction (i.e., numerous planes on the roof), etc. This paper also mentions the canonical example of San Francisco, widely analyzed in the literature, as a case showing the impact of building deorientation on double bounce scattering. This area of interest is also discussed in terms of the impact of SAR data resolution on the detection of specific backscatter mechanisms. The findings of this work are very useful for increasing the awareness of the utilization of classification approaches where only pixels with double bounce backscatter mechanisms are classified as urban areas. Moreover, the article lists factors that should be taken into consideration when performing urban area detection based only on polarimetric data and standard algorithms, such as street and building orientation, building heights, and structures. [ABSTRACT FROM AUTHOR]

Synthetic aperture radar (SAR) plays a vital role in change detection (CD) analysis due to the ability to produce remote sensing images throughout the day, regardless of weather conditions. Nowadays, deep learning methods have gained popularity in multitemporal SAR image CD applications. However, false labels generated during the preclassification stage limit the performance of the CD process. This work employs a fast and robust fuzzy c-means clustering to generate the pseudo-label samples and lightweight spatial pyramid pooling attention network (SPPANet) to detect changes in multitemporal SAR images. The spatial pyramid structure in SPPANet applies adaptive pooling layers to provide better contextual information without incurring computational overhead. The log-ratio operator is used to generate the difference image (DI), and the pseudo-label samples are created from DI. The pseudo-label samples are used to create the training and testing patches. Finally, the trained SPPANet is used to classify testing samples into unchanged and changed classes. The SPPANet achieves an accuracy of 98.70%, 99.06%, 96.40%, and 99.10% for Ottawa, San Francisco, Yellow River, and Farmland datasets, respectively. [ABSTRACT FROM AUTHOR]

What information may be extracted over urban area by means of joint analysis of two-dimensional (2D) and three-dimensional (3D) remote sensing data? We exploit aerial, Synthetic Aperture Radar (SAR) and Laser Induced Detection and Ranging (LIDAR) data to characterize precisely the Presidio area in San Francisco. We discriminate between different objects in the scene using their 2D and 3D characteristics. The final product of the analysis is a set of raster or vector information layers providing land covers, 3D building shapes and Digital Terrain Models (DTMs) of the Presidio. This paper investigates the relative merits of the collected data in retrieving each of these information layers, and examines how automatic algorithms to extract land cover, Digital Terrain Model (DTM) and 3D building shape could be integrated in a processing chain. [ABSTRACT FROM AUTHOR]

In this paper, we propose a new discriminative dictionary learning method based on Riemann geometric perception for polarimetric synthetic aperture radar (PolSAR) image classification. We made an optimization model for geometry-aware discrimination dictionary learning in which the dictionary learning (GADDL) is generalized from Euclidian space to Riemannian manifolds, and dictionary atoms are composed of manifold data. An efficient optimization algorithm based on an alternating direction multiplier method was developed to solve the model. Experiments were implemented on three public datasets: Flevoland-1989, San Francisco and Flevoland-1991. The experimental results show that the proposed method learned a discriminative dictionary with accuracies better those of comparative methods. The convergence of the model and the robustness of the initial dictionary were also verified through experiments. [ABSTRACT FROM AUTHOR]

Polarimetric synthetic aperture radar (PolSAR) images contain useful information, which can lead to extensive land cover interpretation and a variety of output products. In contrast to optical imagery, there are several challenges in extracting beneficial features from PolSAR data. Deep learning (DL) methods can provide solutions to address PolSAR feature extraction challenges. The convolutional neural networks (CNNs) and graph convolutional networks (GCNs) can drive PolSAR image characteristics by deploying kernel abilities in considering neighborhood (local) information and graphs in considering long-range similarities. A novel dual-branch fusion of CNN and mini-GCN is proposed in this study for PolSAR image classification. To fully utilize the PolSAR image capacity, different spatial-based and polarimetric-based features are incorporated into CNN and mini-GCN branches of the proposed model. The performance of the proposed method is verified by comparing the classification results to multiple state-of-the-art approaches on the airborne synthetic aperture radar (AIRSAR) dataset of Flevoland and San Francisco. The proposed approach showed 1.3% and 2.7% improvements in overall accuracy compared to conventional methods with these AIRSAR datasets. Meanwhile, it enhanced its one-branch version by 0.73% and 1.82%. Analyses over Flevoland data further indicated the effectiveness of the dual-branch model using varied training sampling ratios, leading to a promising overall accuracy of 99.9% with a 10% sampling ratio. [ABSTRACT FROM AUTHOR]

Generalized likelihood ratio test (GLRT) is an efficient method to generate difference image (DI) for change detection (CD) using synthetic aperture radar (SAR) images. GLRT is usually applied with a fixed-size moving window to the neighbourhood regions in multitemporal SAR images. The fixed window may be however not optimal for all the pixels under test. To solve disadvantages of the GLRT method, an adaptive circular window is proposed in this work. While the adaptive square window and the best fixed square window achieve an average Kappa coefficient of 83.55% and 82.12%, respectively, the adaptive circular window improves the average Kappa coefficient by 83.65% in six datasets. As another difficulty, speckle noise reduces quality of DI in SAR change detection. Three steps are considered to minimize effects of the speckle noise: (1) a two-stage morphological filter is suggested to reduce the speckle noise; (2) to generate DI, the adaptive circular window for generalized likelihood ratio test (ACWGLRT) is proposed that reduces influence of the speckle noise while preserves the edge details of multitemporal images; and (3) spatial fuzzy c-means (SFCM) is used to reduce effects of the residual speckle-noise during DI classification. The experimental results show superior performance of the proposed change detection method with respect to several competitors. The proposed method has the best Kappa coefficient and percentage correct classification (PCC) in four datasets of Ottawa, San Francisco, Farmland C, and Inland water. [ABSTRACT FROM AUTHOR]

Previous studies have shown that scattering mechanism ambiguity and negative power issues still exist in model-based polarization target decomposition algorithms, even though deorientation processing is implemented. One possible reason for this is that the dynamic range of the model itself is limited and cannot fully satisfy the mixed scenario. To address these problems, we propose a hybrid polarimetric target decomposition algorithm (GRH) with a generalized volume scattering model (GVSM) and a random particle cloud volume scattering model (RPCM). The adaptive volume scattering model used in GRH incorporates GVSM and RPCM to model the volume scattering model of the regions dominated by double-bounce scattering and the surface scattering, respectively, to expand the dynamic range of the model. In addition, GRH selects the volume scattering component between GVSM and RPCM adaptively according to the target dominant scattering mechanism of fully polarimetric synthetic aperture radar (PolSAR) data. The effectiveness of the proposed method was demonstrated using AirSAR dataset over San Francisco. Comparison studies were carried out to test the performance of GRH over several target decomposition algorithms. Experimental results show that the GRH outperforms the algorithms we tested in this study in decomposition accuracy and reduces the number of pixels with negative powers, demonstrating that the GRH can significantly avoid mechanism ambiguity and negative power issues. [ABSTRACT FROM AUTHOR]

Polarimetric SAR (PolSAR) scattering characteristics of imagery are always obtained from the second order moments estimation of multi-polarization data, that is, the estimation of covariance or coherency matrices. Due to the extra-paths that signal reflected from separate scatterers within the resolution cell has to travel, speckle noise always exists in SAR images and has a severe impact on the scattering performance, especially on single look complex images. In order to achieve high accuracy in estimating covariance or coherency matrices, three aspects are taken into consideration: (1) the edges and texture of the scene are distinct after speckle filtering; (2) the statistical characteristic should be similar to the object pixel; and (3) the polarimetric scattering signature should be preserved, in addition to speckle reduction. In this paper, a joint restriction principle is proposed to meet the requirement. Three different restriction principles are introduced to the processing of speckle filtering. First, a new template, which is more suitable for the point or line targets, is designed to ensure the morphological consistency. Then, the extent sigma filter is used to restrict the pixels in the template aforementioned to have an identical statistic characteristic. At last, a polarimetric similarity factor is applied to the same pixels above, to guarantee the similar polarimetric features amongst the optional pixels. This processing procedure is named as speckle filtering with joint restriction principle and the approach is applied to GF-3 polarimetric SAR data acquired in San Francisco, CA, USA. Its effectiveness of keeping the image sharpness and preserving the scattering mechanism as well as speckle reduction is validated by the comparison with boxcar filters and refined Lee filter. [ABSTRACT FROM AUTHOR]

In order to more efficiently mine the features of polarimetric synthetic aperture radar (PolSAR) and establish a more appropriate classification model, this article proposes an improved convolutional autoencoder (ICAE) based on texture feature fusion (TFF-ICAE) for PolSAR terrain classification. First, TFF-ICAE specifically designs a multi-indicator squeeze-and-excitation (MI-SE) block and incorporates it into the CAE network. MI-SE can enhance the essential feature information while suppressing the interference information as much as possible, and it can effectively increase the between-class distance while reducing the within-class distance. Then, TFF-ICAE uses gray level co-occurrence matrix (GLCM) to capture the texture features, and it optimally fuses these texture features and the deep features extracted by ICAE to complete the multilevel feature fusion, elevating the feature representation completeness of the terrain. That is, TFF-ICAE effectively enhances the feature separation capability of different categories while greatly elevating the feature representation completeness. Experiments on the datasets of San Francisco, Oberpfaffenhofen, and Flevoland show that the proposed TFF-ICAE, respectively, achieves overall accuracies of 93.44%, 97.61%, and 97.78%, which are at least 0.92%, 1.52%, and 0.97% higher than other algorithms. Undoubtedly, the superiority of TFF-ICAE is verified on these datasets. [ABSTRACT FROM AUTHOR]

Polarimetric synthetic aperture radar (SAR) has been extensively used in various remote sensing applications. In this article, a unified framework is designed to compare the classification performance of different polarimetric systems, which include quad-polarimetric (QP), compact-polarimetric (CP), and dual-polarimetric (DP). To avoid problems, such as the lack of uniform standards in feature extraction, the classification algorithm is directly based on the statistical characteristics of the coherency/covariance matrix and is implemented by extending the Wishart mixture model (WMM). The GF-3 data set in San Francisco and the AIRSAR agricultural data set in Flevoland are used in the experiment, and the following conclusions are generated. QP can achieve the highest classification accuracy in all classification tasks. When distinguishing three typical classes (water, urban, and vegetation) with very different scattering characteristics, the performance of different polarimetric systems is similar, and QP has only a slight advantage. For classification tasks of different classes with similar scattering characteristics, CP performs better in agricultural scenes, and the overall accuracy (OA) is only reduced by 3%–4% compared with QP. DP performs better in urban scenes, and OA is only reduced by 1%–3% compared with QP. These conclusions can provide guidance for future payloads’ design and the choice of polarimetric operation mode for existing multi-polarimetric SAR systems to achieve the purpose of giving full play to the advantages of different polarimetric systems. [ABSTRACT FROM AUTHOR]

We propose a generic scattering power factorization framework (SPFF) for polarimetric synthetic aperture radar (PolSAR) data to directly obtain N scattering power components along with a residue power component for each pixel. Each scattering power component is factorized into similarity (or dissimilarity) using elementary targets and a generalized volume model. The similarity measure is derived using a geodesic distance between pairs of 4 × 4 real Kennaugh matrices. In standard model-based decomposition schemes, the 3 × 3 Hermitian-positive semi-definite covariance (or coherency) matrix is expressed as a weighted linear combination of scattering targets following a fixed hierarchical process. In contrast, under the proposed framework, a convex splitting of unity is performed to obtain the weights while preserving the dominance of the scattering components. The product of the total power (Span) with these weights provides the nonnegative scattering power components. Furthermore, the framework, along with the geodesic distance (GD) is effectively used to obtain specific roll-invariant parameters such as scattering-type parameter (αGD), helicity parameter (τGD), and purity parameter (PGD). A PGD/αGD unsupervised classification scheme is also proposed for PolSAR images. The SPFF, the roll invariant parameters, and the classification results are assessed using C-band RADARSAT-2 and L-band ALOS-2 images of San Francisco. [ABSTRACT FROM AUTHOR]

Polarimetric synthetic aperture radar (PolSAR) as a typical multi-channel sensor can obtain refined geometrical and geophysical information. In the PolSAR land cover classification task, feature extraction is regarded as a critical step for the final classification. It can employ multi-modal features from the original scattering data, polarimetric target decomposition, and other transformation space. Then, how to efficiently combine multi-modal polarimetric information and extract discriminant features is an important challenge for PolSAR image processing. Graph embedding methods have become a significant technique to deal with feature extraction and dimensionality reduction (DR) problems in recent years. It provides a unified linearization framework in machine learning and other pattern recognition tasks. In this article, an extended tensor embedding framework is introduced to extract the intrinsic features for PolSAR land cover classification. First, each pixel is represented by a feature cube that is constructed by groups of polarimetric scattering signals and target decomposition features in a fixed size patch. Second, an intrinsic matrix is constructed to describe the original geometrical and statistical properties of the samples, and a penalty matrix is designed to represent some constraints. Third, the vector-based algorithms are transformed into tensor space in an unified framework and based on the pair of matrices to obtain the projection matrices in each mode by an iterative optimization process. The effectiveness of the proposed methods is demonstrated on three RADARSAT2 data sets covering the regions of Xi’an, San Francisco, and Flevoland, respectively. The visualization and quantification results show that the proposed method has superiority in land cover classification. [ABSTRACT FROM AUTHOR]

Satellite instruments monitor the Earth's surface day and night, and, as a result, the size of Earth observation (EO) data is dramatically increasing. Machine Learning (ML) techniques are employed routinely to analyze and process these big EO data, and one well-known ML technique is a Support Vector Machine (SVM). An SVM poses a quadratic programming problem, and quantum computers including quantum annealers (QA) as well as gate-based quantum computers promise to solve an SVM more efficiently than a conventional computer; training the SVM by employing a quantum computer/conventional computer represents a quantum SVM (qSVM)/classical SVM (cSVM) application. However, quantum computers cannot tackle many practical EO problems by using a qSVM due to their very low number of input qubits. Hence, we assembled a coreset ("core of a dataset") of given EO data for training a weighted SVM on a small quantum computer, a D-Wave quantum annealer with around 5000 input quantum bits. The coreset is a small, representative weighted subset of an original dataset, and its performance can be analyzed by using the proposed weighted SVM on a small quantum computer in contrast to the original dataset. As practical data, we use synthetic data, Iris data, a Hyperspectral Image (HSI) of Indian Pine, and a Polarimetric Synthetic Aperture Radar (PolSAR) image of San Francisco. We measured the closeness between an original dataset and its coreset by employing a Kullback–Leibler (KL) divergence test, and, in addition, we trained a weighted SVM on our coreset data by using both a D-Wave quantum annealer (D-Wave QA) and a conventional computer. Our findings show that the coreset approximates the original dataset with very small KL divergence (smaller is better), and the weighted qSVM even outperforms the weighted cSVM on the coresets for a few instances of our experiments. As a side result (or a by-product result), we also present our KL divergence findings for demonstrating the closeness between our original data (i.e., our synthetic data, Iris data, hyperspectral image, and PolSAR image) and the assembled coreset. [ABSTRACT FROM AUTHOR]

Polarimetric synthetic aperture radar (PolSAR) image classification plays an important role in various PolSAR image application. And many pixel-wise, region-based classification methods have been proposed for PolSAR images. However, most of the pixel-wise methods can not model local spatial relationship of pixels due to negative effects of speckle noise, and most of the region-based methods fail to figure out the regions with the similar polarimetric features. Considering that color features can provide good visual expression and perform well for image interpretation, in this work, based on the PolSAR pseudo-color image over Pauli decomposition, we propose a supervised PolSAR image classification approach combining learned superpixels and quaternion convolutional neural network (QCNN). First, the PolSAR RGB pseudo-color image is formed under Pauli decomposition. Second, we train QCNN with quaternion PolSAR data converted by RGB channels to extract deep color features and obtain pixel-wise classification map. QCNN treats color channels as a quaternion matrix excavating the relationship among the color channels effectively and avoiding information loss. Third, pixel affinity network (PAN) is utilized to generate the learned superpixels of PolSAR pseudo-color image. The learned superpixels allow the local information exploitation available in the presence of speckle noise. Finally, we fuse the pixel-wise classification result and superpixels to acquire the ultimate pixel-wise PolSAR image classification map. Experiments on three real PolSAR data sets show that the proposed approach can obtain 96.56%, 95.59%, and 92.55% accuracy for Flevoland, San Francisco and Oberpfaffenhofen data set, respectively. And compared with state-of-the-art PolSAR image classification methods, the proposed algorithm can obtained competitive classification results. [ABSTRACT FROM AUTHOR]