Partially supervised classification of optical high spatial resolution images in urban environment using spectral and textural information

Conventional multispectral classification methods exhibit poor performances in the detection of urban objects, in high spatial resolution satellite images. This is because multispectral classification is based on the spectral information of the individual pixels and such spectral information is very moderate in the case of high spatial resolution sensor data. We propose to solve the problem by an integrated approach that considers also the important information contained in the spatial arrangement of pixel intensities by means of two methodological aspects. The first aspect is the data fusion to the detection of urban areas by means of the integration of the spectral information with the spatial information represented by texture features extracted from the grey level co-occurrence matrix (GLCM). The second one is the partially supervised classification that exploits the maps provided by the classification of both spectral and texture information by means of a hierarchical clustering algorithm. The proposed approach has been tested on a multispectral image of the IKONOS MS sensor, at 4-meter spatial resolution, acquired over an urban area in Brazil. A quantitative evaluation of the classification performances, based on the overall and average accuracy values and the separability factor, will be reported.