Thematic Information Extraction from IKONOS Imagery based on Object and Various Features

As to high-resolution remote sensing imagery classification based on object-orientated methodology, the precision is related to feature configuration and classification algorithm. In this paper, the IKONOS multi-spectral image is selected as sample data, and segmentated into many image objects. Firstly, the supervising classification is applied to extract land cover thematic information, based on nearest distance and SVM methods respectively, and the total classification accuracy is analysed and compared with each other. Follow the feature space is changed into different dimensionality, and the relation between feature configuration and classification accuracy is discussed respectively with nearest distance and SVM methods. The results show that the SVM classifier can well relax the relationship between image classification and representative feature configuration, and improved the insufficiency of nearest distance methodology during nonlinearity classification processing.