PERBANDINGAN KLASIFIKASI BERBASIS OBJEK DAN KLASIFIKASI BERBASIS PIKSEL PADA DATA CITRA SATELIT SYNTHETIC APERTURE RADAR UNTUK PEMETAAN LAHAN

Abstract

Utilization of remote sensing data for land mapping has long been developed. In Indonesia, as a tropical region, the cloud becomes a classic problem in observing the Earth’s surface using optical remotely sensor satellite. Synthetic Aperture Radar (SAR) sensor satellite has the ability to penetrate clouds so it can solve cloud cover problems. In this study, the ALOS PALSAR data were used to assess object-based and pixel-based classification techniques. This data was chosen due to its capacity for object recognition based on backscatter characteristics. Object-based classification using the methods of Statistical Region Merging (SRM) for the object segmentation process and Support Vector Machine (SVM) for the classification process, whereas the pixel-based classification using SVM method. In the classification stage, several features of Target Decomposition and Image Decomposition of ALOS PALSAR data have been tested. The accuracy assessment of the classification was conducted using confusion matrix of the Region of Interest (ROI) data using the QuickBird data. Implementation of the object-based classification produced better result comparing to pixel-based classification. The number of optimal features is seven which consisted of three features Freeman Decomposition (Red, Green, Blue), Entropy, Alpha Angle, Anisotropy and Normalized Difference Polarization Index (NDPI). Overall accuracy reached 73.64% for the result of the object-based classification and 62.6% for the pixel-based classification.