R.T. Shin

Identification of terrain cover using the optimum polarimetric classifier

A systematic approach for the identification of terrain media such as vegetation canopy, forest, and snow-covered fields is developed using the optimum polarimetric classifier. The covariance matrices for various terrain cover are computed from theoretical models of random medium by evaluating the scattering matrix elements. The optimal classification scheme makes use of a quadratic distance measure and is applied to classify a vegetation canopy consisting of both trees and grass. Experimentally measured data are used to validate the classification scheme. Analytical and Monte Carlo simulated classification errors using the fully polarimetric feature vector are compared with classification based on single features which include the phase difference between the VV and HH polarization returns. It is shown that the full polarimetric results are optimal and provide better classification performance than single feature measurements.

K-Distribution and Polarimetric Terrain Radar Clutter

A multivariate K- distribution is proposed to model the statistics of fully polarimetric radar data from earth terrain with polarizations HH, HV, VH, and VV. In this approach, correlated polarizations of radar signals, as characterized by a covariance matrix, are treated as the sum of N n- dimensional random vectors; N obeys the negative binomial distribution with a parameter α and mean N. Subsequently, an n- dimensional K- distribution, with either zero or nonzero mean, is developed in the limit of infinite N or illuminated area. The probability density function (PDF) of the K- distributed vector normalized by its Euclidean norm is independent of the parameter α and is the same as that derived from a zero-mean Gaussian-distributed random vector. The above model is well supported by experimental data provided by MIT Lincoln Laboratory and the Jet Propulsion Laboratory in the form of polarimetric measurements.

Theoretical Models For Polarimetric Radar Clutter

The Mueller matrix and polarization covariance matrix are described for polarimetric radar systems. The clutter is modelled by a layer of random permittivity, described by a three-dimensional correlation function, with variance, and horizontal and vertical correlation lengths. This model is applied, using the wave theory with Born approximations carried to the second order, to find the backscattering elements of the polarimetric matrices. It is found that 8 out of 16 elements of the Mueller matrix are identically zero, corresponding to a covariance matrix with four zero elements. Theoretical predictions are matched with experimental data for vegetation fields.

Application of neural networks to radar image classification

A number of methods have been developed to classify ground terrain types from fully polarimetric synthetic aperture radar (SAR) images, and these techniques are often grouped into supervised and unsupervised approaches. Supervised methods have yielded higher accuracy than unsupervised techniques, but suffer from the need for human interaction to determine classes and training regions. In contrast, unsupervised methods determine classes automatically, but generally show limited ability to accurately divide terrain into natural classes. In this paper, a new terrain classification technique is introduced to determine terrain classes in polarimetric SAR images, utilizing unsupervised neural networks to provide automatic classification, and employing an iterative algorithm to improve the performance. Several types of unsupervised neural networks are first applied to the classification of SAR images, and the results are compared to those of more conventional unsupervised methods. Results show that one neural network method-Learning Vector Quantization (LVQ)-outperforms the conventional unsupervised classifiers, but is still inferior to supervised methods. To overcome this poor accuracy, an iterative algorithm is proposed where the SAR image is reclassified using a maximum likelihood (ML) classifier. It is shown that this algorithm converges, and significantly improves classification accuracy. >

Backscattering enhancement of electromagnetic waves from two-dimensional perfectly conducting random rough surfaces: a comparison of Monte Carlo simulations with experimental data

Predictions of an exact numerical model for scattering from a surface randomly rough in two directions are compared with experimental data. The numerical model is based on Monte Carlo simulation using an iterative version of the method of moments known as the sparse-matrix flat-surface iterative approach (SMFSIA). Experimental data is obtained from millimeter wave laboratory experiments in which the bistatic scattering patterns of fabricated surfaces with known statistical parameters were measured. The surfaces studied have both a Gaussian height distribution and correlation function, so that their statistics are characterized by an rms height and correlation length. An rms height of 1 wavelength and correlation lengths ranging from 1.41-3 wavelengths are investigated in this paper, and the phenomenon of backscattering enhancement is observed both in the numerical predictions and experimental data. A comparison of the absolute value of the bistatic scattering coefficient as normalized by the incident power shows the theory and experiment to be in good agreement.

Branching model for vegetation

A branching model is proposed for the remote sensing of vegetation. The frequency and angular responses of a two-scale cylinder cluster are calculated to demonstrate the significance of vegetation architecture. The results indicate that the architecture of vegetation plays an important role in determining the observed coherent effects. A two-scale branching model is implemented for soybean with its internal structure and the resulting clustering effects considered. At the scale of soybean fields, the relative location of soybean plants is described by a pair distribution function. The polarimetric backscattering coefficients are obtained in terms of the scattering properties of soybean plants and the pair distribution function. Theoretical backscattering coefficients evaluated using the hole-correction pair distribution are in good agreement with extensive data from soybean fields. The hole-correction approximation, which prevents two soybean plants from overlapping each other, is more realistic and improves the agreement between the model calculation and experimental data near normal incidence. >

A Finite-Difference Time-Domain Analysis of Wave Scattering from Periodic Surfaces: Oblique Incidence Case

Scattering of electromagnetic waves from periodic surfaces is considered in time-domain for an oblique angle of incidence. The finite-difference time-domain (FDTD) method is used to obtain numerical solutions without resorting to frequency-domain analysis and Fourier transformation. For the application of FDTD method to the oblique incidence case, Maxwells equations are transformed such that the computational domain can be truncated by using periodic boundary conditions. The FDTD method is then used to solve the transformed equations. In solving the transformed equations by the FDTD method, the absorbing boundary conditions are modified and the eigenvalues of the system are determined for the stability analysis. The final results are obtained by using the inverse transformation. Since the transformation is very simple, the computational time is primarily determined by the FDTD solution of the transformed equations. The theoretical results are illustrated by calculating the scattered fields in the computa...

A numerical study of the composite surface model for ocean backscattering

A numerical study of 14-GHz backscattering from ocean-like surfaces, described by a Pierson-Moskowitz spectrum, is presented. Surfaces rough in one and two dimensions are investigated, with Monte Carlo simulations performed efficiently through the use of the canonical-grid expansion in an iterative method of moments. Backscattering cross sections are illustrated for perfectly conducting surfaces at angles from 0 to 60/spl deg/ from normal incidence, and the efficiency of the numerical model enables the composite surface theory to be studied in the microwave frequency range for realistic one-dimensional (1D) surface profiles at low wind speeds (3 m/s). Variations with surface spectrum low-frequency cutoff (ranging over spatial lengths from 21.9 to 4.29 cm) are investigated to obtain an assessment of composite surface model accuracy. The 1D surface results show an increase in hh backscatter returns as surface low-frequency content is increased for incidence angles larger than 30/spl deg/, while /spl nu//spl nu/ returns remain relatively constant, all as predicted by the composite surface model. Similar results are obtained for surfaces rough in two dimensions, although the increased computational complexity allows maximum surface sizes of only 1.37 m to be considered. In addition, cross-polarized cross sections are studied in the two-dimensional (2D) surface case and again found to increase as surface low-frequency content is increased. For both 1D and 2D surfaces, backscattering cross sections within 20/spl deg/ of normal incidence are found to be well matched by both Monte Carlo and analytical physical optics (PO) methods for all low-frequency cutoffs considered, and a comparison of analytical PO and geometrical optics (GO) results indicates an appropriate choice of the cutoff wavenumber in the composite surface model to insure an accurate slope variance for use in GO predictions. This choice of cutoff wavenumber is then applied in the composite surface theory for more realistic ocean spectra and compared with available experimental data.

A numerical study of the composite surface model for ocean scattering

A numerical study of ocean backscattering for surfaces rough in two dimensions is presented. The numerical model is based on Monte Carlo simulation using the sparse matrix-flat surface iterative approach with canonical grid (SMFSIA/CAG), which is a more efficient version of the method of moments that allows large two dimensional surfaces to be treated. Backscattering cross sections are illustrated for perfectly conducting power law spectrum ocean surface models at angles from 0 to 60 degrees from normal incidence. Variations with surface spectrum low frequency cutoff (ranging over spatial lengths from 64 /spl lambda/ to 2 /spl lambda/) are investigated, and demonstrate the accuracy of the composite surface model.

Polarimetric Passive Remote Sensing of Periodic Surfaces

The concept of polarimetry in active remote sensing is extended to passive remote sensing. The potential use of the third and fourth Stokes parameters U and V, which play an important role in polarimetric active remote sensing, is demonstrated for passive remote sensing. It is shown that, by the use of the reciprocity principle, the polarimetric parameters of passive remote sensing can be obtained through the solution of the associated direct scattering problem. These ideas are applied to study polarimetric passive remote sensing of periodic surfaces. The solution of the direct scattering problem is obtained by an integral equation formulation which involves evaluation of periodic Greens functions and normal derivative of those on the surface. Rapid evaluation of the slowly convergent series associated with these functions is observed to be critical for the feasibility of the method. New formulas, which are rapidly convergent, are derived for the calculation of these series. The study has shown that the ...