| Summary: | Compact Polarimetry has gained significant importance in recent years among other earth observation missions due to its low power consumption, simple architecture, and larger swath width. For space-based SAR systems, these parameters are vital to monitor the earth surface continuously for various applications. The main manifestations of Hybrid polarimetry from fully polarimetric systems is transmitting a circular polarization and receiving in linear polarizations. In this paper, we assess the performance of compact polarimetry (hybrid polarimetry) over dual-pol RISAT-1 data for land cover classification over various ground targets using backscattered coefficient values, degree of polarization, and relative phase values. In order to understand the scattering mechanism of the targets, Raney decomposition, Pseudo Three Component decomposition, m-δ and m-χ decompositions were performed on the SAR datasets. The m-χ decomposition has proven to be robust when transmitting component is not perfectly circularly polarized. The support vector machine (SVM) classifier algorithm was used to classify the datasets. Three datasets (viz. RISAT-1 hybrid-pol data, RISAT-1 dual-pol data, and Resourcesat-2 data) were evaluated with SVM classifier and compared using three different kernel parameters, i.e. radial basis function (RBF), Polynomial with degree `2' and Linear. From this paper, it was observed that the SVM with RBF kernel parameter gave highest Overall Accuracy (OA) of 92.34% for hybrid Pol RISAT-1 data. Similarly, the SVM with RBF kernel parameter gave an overall accuracy (OA) of 76.83% for dual-pol RISAT-1 data. SVM has classified the datasets into four classes viz. Urban, Water, Vegetation, and Bare soil. The evaluation of classified datasets were performed using confusion matrix for accuracy assessment. For validating the results, the classified image is compared with the optical imagery of Resourcesat-2 (LISS IV) sensor, Google Earth, and In-situ information that was collected synchronous to the satellite pass on July 5, 2016.
|
|---|