Papers
Papers using this Toolbox
Polarimetric Classification using Expectation Methods
Fully polarimetric NASA/JPL AirSAR data from a rice growing area in Japan is processed using the maximum likelihood merging method. Results from an alpha-entropy analysis are presented but the low polarimetric variation at C-band causes poor class separation. After a likelihood-based segmentation, an iterative expectation method of decomposing the scene into the M most significant classes is discussed which gives a useful classification without a priori knowledge. Although the entire covariance matrix can be incorporated, results obtained using intensity channels only are not significantly inferior. By consideration of the intensity, phase-difference and coherence the relative contribution from each towards class-separation is discussed, and using the sample covariance matrices, this leads to simple measures of separation which justify the observations. Given class covariance matrices, and assuming classes to be fully specified by this, the optimum classification accuracy can now be calculated for an arbitrary number of looks.
Single Look Classification Accuracy for Polarimetric SAR
Closed form expressions are given for the expected single look accuracy of a maximum likelihood classifier based on the Wishart distribution of the polarimetric covariance matrix for SAR data. This can be used to compare the high resolution classification performance of different operating modes for polarimetric systems such as ALOS-PALSAR and Radarsat-2.
Numerical Laplace Analysis of K-Distributed Clutter in Noise
Statistical analysis of the sum of N pulses of K-distributed clutter within noise is considered. A closed form solution is not available and so a high precision numerical scheme is used to invert the closed form Laplace transform of the distribution. Considerable time and accuracy benefits over Monte-Carlo methods are apparent for extreme false alarm rates as low as 1e–10. For typical values, the detection probability is shown to be accurately approximated by using the traditional Swerling models. The required windowed CFAR threshold is calculated and the associated loss compared to fixed threshold detection can be determined. The robustness of the normalised log estimator for the shape parameter is considered.
Other Papers (Radar 2002)
Performance Evaluation of Maximum Likelihood SAR Segmentation for Multi-temporal Rice Crop Mapping
Optimal (Maximum Likelihood) processing for a SAR image comprised of discrete regions of constant radar cross section is now well known. This scheme considerably improves upon windowed and iterative schemes by merging regions on an individual pixel basis. In theory, rigorous expressions for `false alarm rate' can be defined but they are perhaps too sensitive to the underlying assumptions of independence and homogeneity. Images can be visually improved by a restraint on the `surface tension' of the segmented regions but, to avoid subjective judgement, performance is assessed using pixel accuracy ground truth from a rice growing area in central Japan based on multi-temporal, 8m resolution Radarsat data. Gaussian Expectation Maximisation is used to achieve accurate, fully-unsupervised classification of the area without fixed-value thresholding. This large scale backscatter information is then used in a Bayesian merging scheme to give a significant improvement in performance.
Paper (PDF 1.3MB)
Presentation (PPT 2.84MB)
Wavelet Detection of Low Observable Targets Within Sea Clutter
We observe scattering events in the Doppler domain via a Continuous Wavelet Transform (CWT). By minimising the uncertainty in velocity and time, the dominant event in the non-stationary instantaneous Doppler spectrum can be isolated. Considering the returns to be discrete in nature leads to a physically motivated detection statistic termed `persistence' that measures the observed lifetime of the scatterers. A Doppler detection scheme is formulated at low velocities within the clutter spectrum, an area neglected by current Doppler filtering. When operated on real data containing an oil drum target, results appear to be complementary to simple intensity thresholding; this suggests detection is made on secondary effects such as wake disturbance.
Paper (PDF 240KB)
Presentation (PPT 1.51MB)
Statistical Analysis of High Resolution Land Clutter
Results are described from land clutter analysis obtained from the fixed site BYSON radar at Malvern. Digital Terrain Elevation Data is used to determine the imaged regions and the normalized log estimator, U, is measured to investigate the stability of the high resolution heavytailed clutter distribution. A Numerical Laplace inversion scheme is described that can determine the distribution of an arbitrary sum of K-distributed variates within noise. However, without a priori knowledge of the clutter distribution, long-tailed distributions can only be accepted after rejecting the hypothesis of exponential clutter that may contain edges. A likelihood based discrimination test is proposed which can operate simultaneously at the same spatial level as a CFAR window. Results suggest that the majority of the scene is in fact closer to edge corrupted speckle and many `spiky' areas are related to man-made structures.