Mapping of Intra-Urban Land Covers Using Pixel-Based and Object-Based Classifications from Airborne Hyperspectral Imagery

High spectral and spatial resolution hyperspectral data provides great potential to characterize intra-urban land cover classes at material level. In this study, AISA airborne hyperspectral image with 0.68m pixel size were used to classify 12 feature classes. In order to conduct the classification, Support Vector Machine (SVM) classifier was used in pixel-based and object-based approach to test the performance of each method for detailed mapping of urban areas. The result of this study showed that object-based SVM contributes to more accurate characterization of urban land covers from the complex environment. The overall accuracy of pixel-based classification was 74.29%, while object-based classification achieved 88.83%. This study highlights the effectiveness of object-based classification by SVM classifier to map the detailed urban land cover classes.