Abstract:
This work is intended to polarimetric analysis of PALSAR image in order to extract polarimetric SAR observables by means of SAR image processing techniques and classification algorithms. The polarimetric SAR observables possess useful intrinsic information, what makes SAR data (here PALSAR) useful for classification.
In the first part of the thesis, the basic concepts of radar polarimetry and state of the art of its application to remote sensing have been discussed with the aim to define established knowledge and possible future development of research.
The second section of the thesis consists of experimental part, which pursues two tasks. In order to accomplish first task, target decomposition theorems have been applied for extracting all relevant polarimetric parameters. The target decomposition theorems laid down the basis of classification, which is the second task of our experimental work. The purpose of this task is to evaluate possible differences between various SAR observables by performing classification. In this context various classification algorithms have been proposed namely, Parallelepiped, Minimum distance, Maximum likelihood and Decision tree classification. The effect of filtering and ensemble averaging on classification has also been evaluated. The SAR observables have been compared by accuracy estimate related to each classification algorithm. The accuracy estimate plays an important role in giving insight to usefulness of SAR observables by providing error in classification. The research work reaches its goal by comparing classification techniques by accuracy estimate as a key feature. The decision tree classifier is found to be best in terms of overall accuracy
