Surface features’ information extraction from SPOT images with object-oriented classification method

With eCognition software of the object-oriented classification method, different segmentation parameters for each surface features in the images is set in the study area of Zhaidi, Guilin. When initial segmentation parameter is 30, shape is 0.1, color is 0.9, compactness is 0.7 and smoothness is 0.3, vegetation, non-vegetation and water body can be parted accurately. Further separation for vegetation and non-vegetation according to the established classification hierarchy, it is concluded that the results close to ideal if the selected segmentation scale is 80 and 50. Classification to the surface features that have been cut by means of eCognition and manually modification has resulted in relatively high accuracy – the general accuracy up to 96.28% and the Kappa coefficient 95.23%. Contrasting with the result by traditional way, the object-oriented classification method is of greater advantage in classifying high-resolution remote sensing data.