Feature Extraction And Scale Effect Analysis Of High-resolution Image Using Object-oriented Approach

Remote sensing technology offers a very useful tool for understanding the earth, it brings us mass of information on the earth and make the view of human being broaden greatly. With the remote sensing technology developing, the spatial resolution of image has been greatly upgraded from kilometer-level to centimeter-level, at the same time the information of image will be richer and more detailed. In the high-resolution remote sensing images, the features of surface target are not reflected by spectral information merely, but combined with spatial information, therefore in the extraction of high-resolution images, it can not only use pixel spectral information as the rules of classification, but also consider other features such as shape, texture, structure and so on. It is clear that traditional pixel-based method does not apply to high-resolution image information extraction, the emergence of object-oriented classification method solves this problem.In this thesis, we extract vegetation and other typical objects from IKONOS image based on object-oriented classification method.A concept"the optimum overall scale"based on global features of image and corresponding model algorithm is put forward. The classification using traditional pixel-based method is also performed. The experimental result shows that the accuracy of object-oriented multi-scale segmentation method is 87.31%, the optimum overall scale classification has an accuracy of 80.78%, and the pixel-based method is 64.96%. Object-oriented classification significantly has better performance than pixel-based classification, showing superiority in high-resolution images extraction.There are five parts in this thesis:the first chapter talks about the basis of the subject and research background, and analysis the status and trends of research on high-resolution remote sensing image extraction, takes on the structure and organization of the paper. Chapter Two introduces the principles and technology of object-oriented classification comprehensively and systematically, points out the advantages of object-oriented method, and introduces the key technologies, namely multi-segmentation. The third chapter is devoted to the concept of scales under different environment, focuses on the problem of optimal segmentation scale and model. Chapter Four, respectively, experiments with object-oriented and pixel-based mothed using IKONOS multi-spectral image, calculates the two accuracies and makes comparative analysis. Chapter Five gives the evaluation and prospects of object-oriented classification method.