Urban Feature Identification And Classification Based On Hyperion Hyperspectral Remote Sensing Data

With the rapid development of the urbanization, it is of great significance to monitor and grasp the information of the urban environment, improve the urban ecological environment and normalize the urban planning administration. City underlying surfaces, especially the large amount of artificial features in different eras, with different materials and components, have the more complicated light spectrum than natural environment. The abundant spectral information in the hyperspectral data can make up for the deficiency of the spectral resolution with the conventional remote sensing data source; thus, more elaborate identification and classification of the urban targets can be achieved. In this dissertation, the problem of urban targets identification with hyperspectral data is discussed. This dissertation is organized as follows:The research significance and object of using hyperspectral data to deal with the urban studies are addressed first. The general development situation of the hardware of hyperspectral remote sensing is introduced. The research trends of image analysis techniques, such as the atmospheric correction technology, the spectral feature extraction, the image fusion, the ground object identification and classification technique and so on, as well as the applications of the hyperspectral remote sensing in the geological survey, the vegetational analysis, the water environmental monitoring, the agricultural information, the atmospheric environment and some other fields are summarized. The main content and the research framework are also given.Then, the block status quo of this research is introduced. For the Hyperion hyperspectral data, the real reflectivity of the ground object can be obtained after a series of preprocessings, such as the geometric correction, the radiometric calibration, the band selection and the atmospheric correction with the Fast Line-of-sight Atmospheric Analysis of Spectral Hyper-cubes (FLAASH) to eliminate the Smile effect and so on. Taking into account the spectral characteristic in all band range as well as the information and correlation between different bands of some typical ground objects in the research area, the band-resample can be achieved. The band with more preserved information, small correlation and strong divisibility of the ground object can be regarded as the optimum band.On that basis, the development status of the data fusion scheme for the remote sensing image and the merits and demerits of common algorithms are summarized. The Gram-Schimdt (GS) is a doptedto fuse the hypcrspectral data, with the high-resolution SPOT pan chromatic image as the reference images. After image fusion, the spatial resolution is obviously improved with small loss of the spectra] information of the ground objects, and the original shape of the spectrum is maintained.Moreover, the spectral signature of the common ground objects in the urban is analyzed. According to the practical situation of the research area, nine categories of urban ground objects are determined to be the research object through the field investigation and the visual interpretation of the remote sensing image. And then, to deal with the shortage of the endmenber extraction method of the ground objects, the Pixel Purity Index and the Spectral Angle Mapper (SAM) are combined to extract the spectrum of endmenber of the nine categories of ground objects. In addition, the reference spectra library is established as the basis of the following identification and classification of the ground objects.Finally, the SAM and Linear Spectral Unmixing(LSU) are used to identify and classify the hyperspectral image before and after the selection of the optimum band, respectively. The statistic analysis of the image results and the terrain area shows that: the common urban ground objects can be identified precisely with spaceborne hyperspectral data. The identification method is crucial to the results. We also find that the fusion of the hyperspectral image may improve the fineness and the accuracy of the classification results to some extent. When the SAM method is adopted to identify the fused images, the statistical error of the terrain area is only11.61%. Whereas, when the LSU method is adopted to identify the unfused images, the error is as high as65.63%, and the image results is too muddy to identify the distribution and the gather form of different ground objects.