Support Vector Machines: Optimization and Validation for Land Cover Mapping Using Aerial Images and Lidar Data

This work investigates the optimization and validation of Support Vector Machines (SVMs) for land cover classification from multispectral aerial imagery and lidar data. For the optimization step, a new method based on a curve fitting technique was applied to minimize the grid search for the Gaussian Radius Basis Function (RBF) parameters. The validation step was based on two experiments. In the first, four SVM kernel models (Gaussian Radius Basis Function; Linear; Polynomial; and Sigmoid) were tested and compared to each other. In the second, SVMs were compared against two classifiers of different characteristics (the Self-Organizing Map (SOM) and the Classification Trees (CTs)) based on four study areas with different sensor and scene characteristics. The comparison is based on two criteria: overall classification accuracy; and commission and omission errors per class. The results demonstrate: the higher overall classification accuracy; the lower range of commission and omission errors per class of the SVMs as compared to other classifiers.