Support Vector Machine for Urban Land-use Classification Using

Support Vector Machine (SVM), as a powerful statistical learning method, , has been found that its performance on land-use-classification outperform conventional classifiers using multiple features extracted from lidar data and imagery. Therefore, in this paper, we use SVM for urban land-use classification. First, we extract features from lidar data, including multi-return, height texture, intensity; other spectral features can be obtained from imagery, such as red, blue and green bands. Finally, SVM is used to automatically classify buildings, trees, roads and ground from aerial images and lidar point clouds. To meet the objectives, the classified data are compared against reference data that were generated manually and the overall accuracy is calculated. We evaluated the performance of SVMs by comparing with classification results using only lidar data, which shows that the land use classification accuracy was improved considerably by fusing lidar data with multispectral images. Meanwhile, comparative experiments show that the SVM is better than Maximum Likelihood Classifier in urban land-use classification