Impact of DFT Based Speckle Reduction Filter on Classification Accuracy of Synthetic Aperture Radar Images

Abstract

Speckle in SAR images makes it difficult to interpret the image thus reducing the effectiveness of image processing. In remote sensing, image scene classification is an elementary problem which aims to label an image automatically with a specific semantic category. The classification performance of SAR data with speckle is inadequate for many applications. Thus, speckle removal becomes an important pre-processing step for SAR data classification. This study investigates the impact and importance of speckle filtering for classification using ALOS-PALSAR-2 data on San Fran-cisco area. Wishart classifier is chosen for classification of filtered and unfiltered SAR data. The influence of DFT based speckle reduction framework is investigated in terms of classification accuracy.