Unsupervised Classification of Polarimetric SAR Images via Riemannian Sparse Coding

Zhong, Neng; Yang, Wen; Cherian, Anoop; Yang, Xiangli; Xia, Gui-Song; Liao, Mingsheng

doi:10.1109/tgrs.2017.2707243

Cited by 25 publications

(13 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…On the Riemannian manifold, the similarities between points can be measured by the geodesic distance [36]. The widely used geodesic distances include the affine invariant Riemannian metric (AIRM) [36], the log-Euclidean distance (LED) [48], and the Bartlett distance [49]. Due to the eigenvalue decomposition in the equation, AIRM has high computational complexity [27].…”

Section: Composite Kernel (Ck) Constructionmentioning

confidence: 99%

“…Although these methods have achieved remarkable breakthroughs, the demands for a large number of labeled samples and their sensitivity to training parameters remain to be solved [12,35]. Since the PolSAR matrices form a Riemannian manifold instead of Euclidean space [36], other classification methods based on CV matrices utilize the similarities between PolSAR matrix samples in the manifold [27,28,36]. Among these methods, representation-based classification methods [27,28] are flexible and can be applied to different polarized SAR datasets without certain distribution assumptions and training processes [27].…”

Section: Introductionmentioning

confidence: 99%

“…To suppress the interference of speckles, the consideration of the spatial correlations contained in PolSAR image is one of the most commonly used and effective methods [39]. Hence, other above-mentioned methods use image patches [2,28,[31][32][33][34] or superpixels [36] to incorporate the spatial information into PolSAR image classification. In this way, improved and smoother classification results can be achieved.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

PolSAR Image Classification Using a Superpixel-Based Composite Kernel and Elastic Net

Cao

et al. 2021

Remote Sensing

View full text Add to dashboard Cite

The presence of speckles and the absence of discriminative features make it difficult for the pixel-level polarimetric synthetic aperture radar (PolSAR) image classification to achieve more accurate and coherent interpretation results, especially in the case of limited available training samples. To this end, this paper presents a composite kernel-based elastic net classifier (CK-ENC) for better PolSAR image classification. First, based on superpixel segmentation of different scales, three types of features are extracted to consider more discriminative information, thereby effectively suppressing the interference of speckles and achieving better target contour preservation. Then, a composite kernel (CK) is constructed to map these features and effectively implement feature fusion under the kernel framework. The CK exploits the correlation and diversity between different features to improve the representation and discrimination capabilities of features. Finally, an ENC integrated with CK (CK-ENC) is proposed to achieve better PolSAR image classification performance with limited training samples. Experimental results on airborne and spaceborne PolSAR datasets demonstrate that the proposed CK-ENC can achieve better visual coherence and yield higher classification accuracies than other state-of-art methods, especially in the case of limited training samples.

show abstract

Section: Composite Kernel (Ck) Constructionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

PolSAR Image Classification Using a Superpixel-Based Composite Kernel and Elastic Net

Cao

et al. 2021

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Among the utilization of PolSAR data, land cover classification is much in demand. A number of PolSAR classification methods, including target decomposition [4], [5] multiple statistical distributions [6], [7] and sparse representation [8], [9] have been developed in recent years. Generally, there are two main steps for PolSAR image classification: feature extraction and representation, and classifier designing and optimization.…”

Section: Introductionmentioning

confidence: 99%

Three-Channel Convolutional Neural Network for Polarimetric SAR Images Classification

Hua

Xie

Jin

2020

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Terrain classifications is an important topic in polarimetric synthetic aperture radar (PolSAR) image processing and interpretation. A novel PolSAR classification method based on three-channel convolutional neural network (Tc-CNN) is proposed and this method can effectively take the advantage of unlabeled samples to improve the performance of classification with a small number of labeled samples. Several strategies are included in the proposed method. (1) In order to take the advantage of unlabeled samples, a data enhancement method based on neighborhood nearest neighbor propagation (N3P) method is proposed to enlarge the number of labeled samples. (2) To increase the role of central pixel in CNN classification based on pixel, a spatial weighted method is proposed to increase the weight of central pixel features and weak the weight of other types of pixel features. (3) A specific deep model for PolSAR image classification (named as Tc-CNN) is proposed, which can obtain more scale and deep polarization information to improve the classification results. Experimental results show that the proposed method achieves a much better performance than existing classification methods when the number of labeled samples is few. Index Terms-convolutional neural network (CNN); polarimetric SAR; terrain classification; three-channel convolutional neural network (Tc-CNN)

show abstract

“…grouped into two main categories, namely unsupervised and supervised algorithms, and have been extensively investigated [9], [12], [15]- [21]. The former consists of clustering image pixels by means of common characteristics/features and occurs in an automatic way without any kind of aid from the user.…”

Section: Introductionmentioning

confidence: 99%

An Eigenvalue-Based Approach for Structure Classification in Polarimetric SAR Images

Biondi

Clemente

Orlando

2020

IEEE Geosci. Remote Sensing Lett.

View full text Add to dashboard Cite

In this paper, we design a novel unsupervised architecture for automatic classification of the dominant polarization in polarimetric SAR images. To this end, we leverage the ideas developed in [1] and suitably exploit them to build a decision logic capable of recognizing the dominant scattering mechanism which characterizes the pixel under test. Specifically, we combine the original data to generate three different sets of reducedsize vectors, which feed a dominant eigenvalues classifier based upon the Model Order Selection rules. Then, the outputs of the latter classification schemes are exploited to infer, according to a specific criterion, the dominant polarization. The performance analysis is conducted on measured data and points out the effectiveness of the newly proposed classification architecture also showing that information about the dominant polarization can be representative of the type of structure which gives raise to the dominant backscattering mechanism.

show abstract

Unsupervised Classification of Polarimetric SAR Images via Riemannian Sparse Coding

Cited by 25 publications

References 37 publications

PolSAR Image Classification Using a Superpixel-Based Composite Kernel and Elastic Net

PolSAR Image Classification Using a Superpixel-Based Composite Kernel and Elastic Net

Three-Channel Convolutional Neural Network for Polarimetric SAR Images Classification

An Eigenvalue-Based Approach for Structure Classification in Polarimetric SAR Images

Contact Info

Product

Resources

About