The H/A/Alpha Polarimetric Decomposition Theorem and Complex Wishart Distribution for Snow Cover Monitoring

This study discusses the capability of full polarimetric L-band ALOS PALSAR data for snow classification. In this study, Polarimetric decomposition and the complex Wishart classifier are applied on ALOS-PALSAR data. Optical (ALOS-AVNIR) data within six days difference was used for visual interpretation of snow and non-snow classes. Application of Entropy-Anistropy-Alpha-Wishart classifier for training samples gives better classification results. To reduce speckle effects and to improve classification results, the refined Lee filter was applied on the covariance matrix several times, each time increasing the size of the moving window. Over all classification accuracy were observed 75.61%, 91.46%, 94.91%, 96.19%, 97.16% and 98.37% using different window sizes 1times1 (without filtered image), 3times3, 5times5, 7times7, 9times9 and 11times11 respectively for refined lee speckle filter. It is observed that classification accuracy increases as size of the filter window increases for speckle reduction. Polarization signatures of various features have also been generated using polarization synthesis techniques and signatures are represented in 3-D plot.