Study On Urban Man-made Objects Extraction Methods In High Resolution Remote Sensing Satellite Images

Recent years, with the improvement of spatial resolution of the satellite sensors, high resolution remote sensing images have become one of the most important data sources for urban information extraction. For supplying technical support to collecting and updating urban data, urban man-made objects (roads and buildings) extraction methods in high resolution remote sensing satellite images are studied in this paper. And the work in this paper focuses on following aspects: 1 Considering the characteristics of high resolution images, the technical difficulties of extraction from high resolution remote sensing image are analyzed. And the problems of the traditional segmentation methods based on pixel spectral attribute are given. The advantages of the segmentation methods based on objects' several kinds of attributes are illustrated. 2.The Quickbird remote sensing images of Yancheng are selected as the main research images. Based on these images, the features of buildings and roads are analyzed in different spectral images. And according to those features, single-band remote sensing image or multi-band fusion image which is best for objects extraction is selected as the further processing image. 3.0bject Oriented Multi-scale segmentation method is further studied. And the optimal scale parameters of roads and buildings are obtained respectively through the experiments. Then considering the features of roads and buildings, the optimal segmentation parameters of roads and buildings are obtained in corresponding image level through the experiments. 4.Mean Shift segmentation based on confidence is studied. And the optimal segmentation parameters of roads and buildings in this method are obtained by experiments. 5.The initial contours of man-made objects are obtained by two methods mentioned above through experiments. The result shows that compared to Object Oriented Multi-scale segmentation method, Mean Shift segmentation based on confidence has more advantages in high resolution remote sensing image segmentation. Therefore, Mean Shift segmentation based on confidence is available and effective in extracting the urban man-made objects from high resolution remote sensing satellite images.