Urban Object Classification Based On High Spatial Resolution Hyperspectral Remote Sensing Data

With the rapid economic development and the urbanization deepening, the spatial distribution and functional organization of the city is undergoing significant change. Therefore, fast and accurate access to the city feature information is crucial to the complexity of urban systems management, evaluation and monitoring of the urban environment, urban planning and other issues for future development. The traditional ground-based survey instruments are time-consuming, laborious, and difficult to update the data. The effect is not very satisfactory. Remote sensing technology has the advantage, that, it can simultaneously acquire a large area of the image, with wide coverage, high update speed. Different surface features have different spectral reflectance characteristics, which is the basis of monitoring urban development by using remote sensing data. Hyperspectral data provides a rich spectral information with Images and spectra exist together " feature. It is suitable for urban information extraction which has spatial details and diversity rich features.There are big differences between reflectance spectra because the improvement of the spatial resolution of the remote sensing data and the influence of various noise. The reflectance spectra volatility becomes larger even when they belong to the same category. Some traditional classification methods of hyperspectral remote sensing do not applied well for the high spatial resolution hyperspectral remote sensing data classification. This paper aims at the goal, that the city feature can be classified precisely by using high spatial resolution hyperspectral remote sensing data.In this this paper, the background of hyperspectral classification are firstly described, the development of hyperspectral remote sensing technology are reviewed and the research status of hyperspectral remote sensing classification techniques are summarized. Secondly, to establish the basic of the classification research, the characteristics of the city type feature are analyzed, and an urban feature type spectral library based on field survey figures reflectance spectral are designed and implemented. Then a classification scheme that combine the spectral feature and spatial feature are designed as follow:(1) high spatial resolution hyperspectral data is divided in space;(2) all spectral elements contained in each object(divided areas) are calculated the average based on the segmentation results;(3) according to each object reflectance spectral curve, spectral database are used to identify or classify ground object by matching method. At last, high resolution spectral ASIA data that acquired in an airborne remote sending experiment are used to test the classification scheme, before which the remote sensing data are preprocessed including radiometric calibration, atmospheric correction and geometric correction. The results show that the classification method used can effectively improve the urban object classification accuracy, in particular, can effectively reduce the scatter, and provide a good reference for information extraction and urban planning studies.