DIMENSIONALITY REDUCTION BASED ON MAXIMUM MARGIN LOCAL SCALING CUT FOR POLSAR IMAGE CLASSIFICATION

摘要

Thanks to the high radar resolution, land-cover classification is one of the most significant applications for PolSAR images. Apart from the selection of classifiers, the suitable features also have a great influence on classification results and efficiency. This paper proposes a novel supervised dimensionality reduction method called Maximum Margin Local Scaling Cut (MMLSC) based on Maximum Margin Criterion (MMC) and Graph Cut Theory. Experiments demonstrate this algorithm gets a better result than some classical algorithms.