Multiple Crop Classification Using Various Support Vector Machine Kernel Functions

摘要

This study was carried out with techniques of Remote Sensing (RS) based crop discrimination and area estimation with single date approach. Several kernel functions are employed and compared in this study for mapping the input space with including linear, sigmoid, and polynomial and Radial Basis Function (RBF). The present study highlights the advantages of Remote Sensing (RS) and Geographic Information System (GIS) techniques for analyzing the land use/land cover mapping for Aurangabad region of Maharashtra, India. Single date, cloud free IRS-Resourcesat-1 LISS-III data was used for further classification on training set for supervised classification. ENVI 4.4 is used for image analysis and interpretation. The experimental tests show that system is achieved 94.82% using SVM with kernel functions including Polynomial kernel function compared with Radial Basis Function, Sigmoid and linear kernel. The Overall Accuracy (OA) to up to 5.17% in comparison to using sigmoid kernel function, and up to 3.45% in comparison to a 3rd degree polynomial kernel function and RBF with 200 as a penalty parameter.