Thematic maps representing the characteristics of the Earth’s surface have been widely used as a primary input in many land related studies. Classification of remotely sensed images is an effective way to produce these maps. The value of the map is a function of the accuracy of the classification. Selection of proper size of samples and classification method are important factors which govern accuracy of thematic maps. In the present study, training data sets of various sizes are used to investigate their effects on the classification accuracy. Two investigations have been carried out. The first one makes use of equal sizes of training data for the classification of 0.6 meter spatial resolution QuickBird-2 satellite image. The second experiment allocates higher sampling fraction for the classes of interest while reducing the number of samples in the less important categories. Six supervised classification methods with different characteristics are applied to produce land use/land cover thematic map of the study area. The classifiers used in the study include: Parallelepiped, Minimum distance, Mahalanobis distance, Maximum likelihood, Neural Net work and Support Vector Machine (SVM). After certain fraction of sample size, the classification accuracy showed downward trend with the increasing number of training pixels. In the case of limited number of training pixels, SVM and maximum likelihood classifiers showed higher classification accuracies than the rest of classifiers. In the case of proportional training size sample, the overall accuracies of all classifiers have been reduced as compared with the first experiment except for SVM classifier. |
