Classification of Objects from High Resolution Remote Sensing Images using eCognition

High resolution satellite images offer rich abundance information of the earth surface including spatial, spectral and contextual information. In order to extract the information from these high resolution images, we need to utilize the spatial and contextual information of an object and its surroundings. If pixel based approaches are applied to extract information from such remotely sensed data, only spectral information is used. Thereby, Pixel based approaches can’t satisfy high resolution satellite image’s classification and the information extraction is based exclusively on the gray level thresholding methods so this produced the large data redundancy. To overcome this situation an object-oriented approach is implemented. This paper demonstrated the concept of object oriented information by using eCognition software, allows the classification of remotely-sensed data based on different object features, such as spatial, spectral, and contextual information. Thus with the object based approach, information is extracted on the basis of meaningful image objects rather than individual gray values of pixels. The test area has different discrimination, assigned to different classes by this approach: Vegetation, Shadow, Artificial patches, Soil patches, Roads and Buildings. The Multiresolution segmentation and the nearest neighbor (NN) classification approaches are used and overall accuracy is assessed.