Unsupervised classification using polarimetric decomposition and complex Wishart classifier

Abstract

In this paper, we propose a new method for unsupervised classification of terrain types and man-made objects using polarimetric S A R data. This technique is a combination of the unsupervised classification based on the polarimetric target decomposition (Cloude and Pottier, 1997) and the maximum likelihood classifier based on the complex Wishart distribution (Lee et al., 1994). The advantage of this approach is that clusters may be identified by the scattering mechanisms from the target decomposition. The effectiveness of this algorithm is demonstrated using JPUAIRSAR and SIR-C polarimetric SAR images. MTRODUCTION Terrain classification is an important polarimetric SAR (POLSAR) application.Many algorithms have been developed for supervised and unsupervised terrain classification. In principle, training sets for each class are selected manually in the supervised classification. This manual technique could be difficult to implement due to the high dimensionality of POLSAR imagery. Unsupervised classification automatically finds clusters based on a certain criterion. In most algorithms, clusters have to be interpreted manually for terrain types to be useful.Cloude and Pottier [l] proposed an unsupervised classification algorithm based on their target decomposition theory. The medium's scattering mechanisms, characterized by entropy (H) and a angle are used for classification. The H and a plane is divided into 8 zones. The physical scattering characteristic associated with each zone provides information for terrain type assignment. This distinctive advantage, unfortunately, is offset by preset zone boundaries in H and a plane. Clusters may fall on the boundary or more than one class may be enclosed in one zone. Furthermore, magnitudes are not used in the classification.Lee et a1 [2] developed a supervised algorithm based on the complex Wishart distribution for the polarimetric covariance matrix. Training sets have to be selected in advance.We propose to use the Cloude and Pottier's method to initially classify the polarimetric SAR image. The initial classification map serves as training sets for classification based on the Wishart distribution. The classified results are then used as training sets for the next iteration. Significant improvement in each iteration has been observed. The iteration ends, when the number of pixels switching classes becomes smaller than a predetermined number. We have observed that the class centers in the H -E plane are shifted after the first and subsequent iterations. The class boundaries in the H -a plane become indistinct with considerable overlap. However, the final class centers in H -E plane are useful for interpretation of terrain types. The advantage of this combined method is in its effectiveness in automated classification, and in providing interpretation based on scattering mechanism for each class. Classification Based on Target DecompositionCloude and Pottier's decomposition is based on the eigenvalue analysis of the coherency matrix, T, which is formed by the Pauli base...

Assessment of L-Band Sar Data at Different Polarization Combinations for Crop and Other Landuse Classification

“…2(b)). The details of these classification algorithms can be found in literature [6,[10][11][12][13][14][15]18]. We have however discussed the algorithm and methodology of the present classification in brief.…”

“…The SAR polarimetric covariance matrix obeys the statistics of a complex multivariate Wishart distribution [12][13][14]. The probability density function of the n-look covariance matrix, Z, is…”

Polarimetric SAR Image Classification Using Radial Basis Function Neural Network

“…Over the past decade, there has been extensive research in the area of the segmentation and classification of polarimetric SAR data. In the literature, the classification algorithms for polarimetric SAR can be divided into three main classes: 1) classification based on physical scattering mechanisms inherent in data [1,2], 2) classification based on statistical characteristics of data [3,4] and 3) classification based on image processing techniques [5,6]. Additionally, there has been several works using some combinations of the above classification approaches [3,1].…”

“…In the literature, the classification algorithms for polarimetric SAR can be divided into three main classes: 1) classification based on physical scattering mechanisms inherent in data [1,2], 2) classification based on statistical characteristics of data [3,4] and 3) classification based on image processing techniques [5,6]. Additionally, there has been several works using some combinations of the above classification approaches [3,1]. While these approaches to the polarimetric SAR classification problem can be based on either supervised or unsupervised methods, their performance and suitability usually depend on applications and the availability of ground truth.…”

An Improved Sar Radiometric Terrain Correction Method and Its Application in Polarimetric Sar Terrain Effect Reduction

“…These applications include land use/land cover mapping [1][2][3][4][5][6][7][8][9], change detection, hazard monitoring and damage assessment, surface geophysical parameters retrieval [10], biomass and forest height estimation, etc. [11].…”