Research On Key Technology Of Hyper-Spectral Remote Sensing Image Processing

With development of traditional multi-spectral infrared sensing technology, hyper-spectral remote sensing as one of new remote sensing technologies is being rapidly developed in the basis of electromagnetic spectrum, geographic information system, electronic technology, computer technology, and aerospace technology. Due to high spectrum resolution from hyper-spectral image data, rich information of field spectra has been gradually paid attention. In the field of traditional remote sensing image processing, there are some matured methods developed. In contrast, there are much more difficulties in hyper-spectral image data processing because much higher spectral data dimensionalities and much more image data for higher resolution need to be dealt with. Consequently, there are some limitations for the applications on hyper-spectral remote sensing image with normal multi-spectral image processing methods. It is obvious that there are theoretical meanings and application values to develop much more efficient analysis and processing for hyper-spectral image in order to fully use potential advantage of hyper-spectral image data. Furthermore, data classification processing technology of hyper-spectral image is one of hot studies in hyper-spectral image processing. In this thesis, band selection method of hyper-spectral image data has been studied using conventional multi-spectral remote sensing image processing together with lowering dimensionalities and classification of hyper-spectral image data.Firstly, to select band combination of high spectral dimension and lower data dimensionalities in hyper-spectral image, this thesis gives subspace-decomposition adaptive band selection (SABS) method of hyper-spectral image using subspace decomposition technique. With this method, selected bands have reasonable distributions over whole spectral space, which not only lowers data dimensionalities of hyper-spectral image and reduces correlation of adjacent spectral bands but keep local classification characterization also. With highly correlated characterization of various bands in hyper-spectral image, necessity and practicality for band selection have been discussed. In band selection method, after subspace decomposition for data source, adaptive selected bands within...