Research On Technologies Of Hyperspectral Data Processing

Based on the data source produced by the first national spacebornehyperspectral imaging spectrometer in business operation, this dissertation mainlystudies several key problems in the hyperspectral data processing, and then proposesassociated solutions and methodologies, which could be bases for hyperspectral dataprocessing and application.The contents mainly include the following contexts:1. The remote sensing technologies for hyperspectral are summarized. Theconcepts and applications of hyperspectral remote sensing, classification ofimaging spectrometers and technologies of data processing are brieflydescribed.2. As it is partial to assess destriping algorithm only from the aspect ofmaintenance of image information currently, a new criterion is proposed toassess the ability of destriping algorithm to maintain the spectral information.Based on analyzing the generating mechanism of strips in the data ofinterference imaging spectrometers, a method for image processing isproposed to eliminate the strip in the data of HJ-1A HSI, which producedbetter results in the maintenance of image information and spectralinformation. The differences are compared between calibration methods ofHJ-1A HSI in library and On-Orbit, and than a new method for absoluteradiance calibration is presented to destrip the image of push-broominterference imaging spectrometers.3. The performance of FLAASH and QUAC to correct atmosphere in HJ-1AHSI data is analyzed, and then a quick atmospheric correction method basedon the characteristic spectrum of thick clouds is presented. With qualifiedaccuracy in atmospheric correction, the method acquires much higher speedthan QUAC. 4. Based on the formula of spectral distance, a variation trend of spectraldistance with increasing bands is deduced. The trend is used to optimizebands selection by controlling the change direction of the average spectraldistances among materials. An optimization method for band selection isproposed in the spectral discrimination.5. In order to correct the errors in classification produced by the uncertainty ofspectral information, an algorithm of supervised spectral classification basedon the spatial information is presented. This algorithm produced better resultsthan that of the SAM algorithm as used to classify the spectral data of HJ-1AHSI. An improved algorithm of unsupervised spectral classification based onspatial information is used as a preprocessing of endmember extraction,where the endmember is extracted from one pixel classes rather than from allthe pixels in the image, which greatly improve the speed of algorithm.