Classification of Earth terrain using polarimetric synthetic aperture radar images

Abstract

Supervised and unsupervised classification procedures are developed and applied to synthetic aperture radar (SAR) polarimetric images in order to identify their various Earth terrain components. For supervised classification processing, the Bayes technique is used to classify fully polarimetric and normalized polarimetric SAR data. Simpler polarimetric discriminates, such as the absolute and normalized magnitude response of the individual receiver channel returns, in addition to the phase difference between the receiver channels, are also considered. Another processing algorithm, based on comparing general properties of the Stokes parameters of the scattered wave to that of simple scattering models, is also discussed. This algorithm, which is an unsupervised technique, classifies terrain elements based on the relationship between the orientation angle and handedness of the transmitting and receiving polarization states. These classification procedures have been applied to San Francisco Bay and Traverse City SAR images, supplied by the Jet Propulsion Laboratory. It is shown that supervised classification yields the best overall performance when accurate classifier training data are used, whereas unsupervised classification is applicable when training data are not available. INTRODUCTIONClassification of Earth terrain within an image is one of the many important applications of polarimetric data. A systematic classification procedure will place the classification process on a more quantitative level and reduce the amount of photointerpretation necessary [Thompson, 1986;Evans et al., 1986]. Single feature and multifrequency classification has been used in the past, but classification can now be applied to fully polarimetric data which have become available due to recent developments in radar technology [Stovall, 1978;Novak and Sechtin, 1986]. It has been shown that Bayes classification using fully polarimetric data yields optimal results as compared to classification performance using any subset of the complete polarimetry [Kong et al., 1988].In some cases, the absolute magnitude and phase of the radar return are not reliable features for data classification purposes. This is due to the fact that radar system calibration procedures vary in accuracy as well as the fact that they cannot account for attenuation and phase shifts caused by atmospheric distortions. Normalization schemes [Yueh et al., 1988], which preserve only the relative components of the returns, were applied to radar polarimetry in order to circumvent this problem. Previously, normalized polarimettic classification schemes were implemented by assuming a multivariate Gaussian distribution for normalized data [Kriegler et al., 1971; Smedes et al., 1971]. However, this technique yields inconsistent results since the probability of error, as well as classification performance, becomes a function of the particular normalization scheme selected. Therefore the optimal normalized classification algorithm will be employed in which the probability density funct...

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.…”

Three-dimensional X-band SAR imaging of a small conifer tree

“…SAR oOE ers an important tool in the monitoring of forestry and agriculture (Lim et al 1989, Dobson et al 1995, which is the particular concern of this work. However, eOE ective use of the data requires proper understanding of the interaction between the radar wave and vegetation.…”

Application of Mellin-Kind Statistics to Polarimetric <inline-formula> <tex-math notation="TeX">${\cal G}$</tex-math></inline-formula> Distribution for SAR Data

“…The gaussian counterpart for the multilook PolSAR case is the matrix-variate scaled complex Wishart distribution [12]. Both these models have been experimentally verified on real PolSAR data [13], [14]. In the context of scalar texture product model, different distributions for the texture random variable will result in different expressions for the resulting compound distribution.…”