Feature selection and classification for urban data using improved F-score with Support Vector Machine

Remote sensing images are relevant materials for observation and thematic mapping by multispectral and multi-textural classification. In this paper, we propose classification of urban data with high spectral and spatial resolution. The approach is based on building Differential Morphological Profile (DMP) and then classifying each pixel using Support Vector Machines (SVM) classifier. The DMP is used for defining a set of features for every structure. Then, the DMP images are used as input in the SVM classifier in order to assign each structure to one of the classes. Since the set of DMP images is often redundant, a feature selection step is performed aiming at reducing the dimensionality of the feature set before applying the SVM classifier. The proposed selection is based on the use of the improved F-score technique with SVM for selecting the most relevant feature DMP images and classifying the urban structures. The method has been applied on panchromatic IKONOS data from urban areas to classify urban structures. The obtained results for approach based on use of DMP with feature reduction show the effective use of DMP with feature reduction compared to those obtained without any feature reduction.