LAUSR

ABSTRACT The present work aims at classifying the natural and man-made objects by fusing features of coarse resolution hyperspectral (1 m) LWIR and fine resolution (20 cm) RGB data. The classified results comprise of five classes namely, road, trees, building, vegetation and soil. The methodology includes extraction of spatial and spectral features to obtain the knowledge base for various classes. Besides vegetation index and morphological building index, the features extracted also include the textural features to obtain the database on spatial values for all the different classes. After extracting these features, bounding boxes have been generated to have appreciable information on the edges of different classes. Finally, connected component analysis has been used for segmentation of classes. The training and testing samples are generated through the knowledge base of connected components which is uniquely fed to Support Vector Machine (SVM) classifier for classification purpose. The classified results indicate definite improvement in object-based classification using multisensor data.