| This dissertation develops a novel method for classifying short vegetation using synthetic aperature radar. Underlying this problem is the problem of distinguishing fine differences in geometry and dielectric constant based on the polarization, magnitude, and phase response of electromagnetic reflections.; We take a statistical approach, beginning by developing a novel method of solving the entropy-moment problem for maximum-entropy density estimation. The technique works in higher dimensions and is computationally efficient and numerically accurate.; We combined this density estimation technique with Bayesian methods and hierarchical classification techniques to come up with what we call a Bayesian-Hierarchical classifier for short vegetation.; The overall accuracy, when applying this technique the data we have available is 93% for the classification of short vegetation. We have used a large data set, with ground-truth information, and independent training and testing data sets. These points are significant, because if any of these factors is not part of the analysis, it is possible to obtain a high classification accuracies that do not reflect the actual performance of the classifier when used with real data.; Finally, we discuss and demonstrate the application of the aforementioned statistical techniques above to represent the non-stationarity in a vegetation target. This representation is applied to perform a simple form of biophysical parameter estimation. We further demonstrate the application of the classification principles to point target detection, applying a deconvolution technique that proves useful for modeling point target statistics in a vegetated background. |
