Abstract
The rapid advancements in technology have facilitated easy availability of multisensor and multiresolution remote
sensing data. Multisensor, multiresolution data contain complementary information and fusion of such data may result in
application dependent significant information which may otherwise remain trapped within. The present work aims at
improving classification by fusing features of coarse resolution hyperspectral (1 m) LWIR and fine resolution (20 cm)
RGB data. The classification map comprises of eight classes. The class names are Road, Trees, Red Roof, Grey Roof,
Concrete Roof, Vegetation, bare Soil and Unclassified. The processing methodology for hyperspectral LWIR data
comprises of dimensionality reduction, resampling of data by interpolation technique for registering the two images at
same spatial resolution, extraction of the spatial features to improve classification accuracy. In the case of fine resolution
RGB data, the vegetation index is computed for classifying the vegetation class and the morphological building index is
calculated for buildings. In order to extract the textural features, occurrence and co-occurence statistics is considered and
the features will be extracted from all the three bands of RGB data. After extracting the features, Support Vector
Machine (SVMs) has been used for training and classification. To increase the classification accuracy, post processing
steps like removal of any spurious noise such as salt and pepper noise is done which is followed by filtering process by
majority voting within the objects for better object classification.
