Water-body types classification using Radarsat-2 fully polarimetric SAR data

Synthetic Aperture Radar (SAR) data have long been used in water management. To the best of our knowledge, previous publications mainly focus on precise water-body areas extraction and flood monitoring. The main purpose of this paper is to classify water-body into different types according to its function. Firstly, the water-body areas are extracted using Wishart-ML classifier and the false alarms from built-up areas are removed by spatial contextural information. Afterwards, each region in water-body extraction result is regarded as an object and its shapes and polarimetric features are obtained. Random forest (RF) classifier is used in the classification. The Radarsat-2 fully polarimetric (FP) SAR data acquired over Suzhou city, China, are used in our experiments. In the study site, the water-body is divided into three categories: lakes, canals and ponds. Along with them, roads and grasslands are also considered in classification due to their similar properties to water-body in PolSAR data. The overall accuracies of the experimental results reach 89.40% and 96.22% in object-level and pixel-level, which demonstrate the effectiveness of the proposed method and Radarsat-2 FP data in water-body types identification.