Researches On Several Critical Problems Of Hyperspectral Remote Sensing Image Processing And Geologic Application

As one of the state-of-the-art remote sensing sciences and technologies, the depthand the breadth of hyperspectral remote sensing science reaearch and technologyapplication comes to be expanded. This paper focuses on four problems in twodomains of hyperspectral remote sensing image processing and its geologicapplication. The first is the spectral unmixing problem, the second is thehigh-performance computation of hyperspectral remote sensing image processing, thethird is the technical system of hyperspectral remote sensing used in minernaltridimensional exploration, and the fourth is the altered information extraction ofairborne hyperspectral remote sensing based on gas and oil macro seepage.As to the first problem, endmember extraction algorithm, as the cirtical spectralunmixing algorithm of hyperspectral image process, is discussed. And a newendmember extraction algorithm classification method is proposed based on thedifferent design mechanism, that is, the geometric design based on distance, volume,optimization, transposition and projection, and the statistical theory and signaldetection theory design, and the combination of space and spectral information design.At the same time, a new spectral unmixing algorithm is proposed based on that thecomponent of the unconstrained least squares solution vector of the linear spectralmixing model can be defined as the distance from the pixel to the endmember.Because of the mass data property of the engineering application of hyperspectralremote sensing, the parallel computation platform architecture and its programmingcircumstance has been summarized. And the details of cluster based on MPIprogramming and CPU+GPU based on CUDA programming related to hyperspectralremote sensing image parallel process have discussed as two cases. One case is the convirance matrix parallel computation based on MPI, and the other case is the SCMmapping algorithm parallel computation based on GPU.According to the mineral exploration practice of hyperspectral remote sensing,the contents of rock-minernal spectroscopy based on spectrometry, classification andrefined identification methods of rock-minernal, geologic origin circumstanceexploration, and metallogenic prognosis have been discussed. And a technical systemof hyperspectral remote sensing used in minernal tridimensional exploration based onthe hyperspectral remote sensing data processing flow during its application ofresource exploration.Another hyperspectral remote sensing application discussed in this paper is basedon the gas and oil macro seepage. The airborne hyperspectral SASI data is processedto extract hydrocarbon, altered carbonate and altered clay. The SASI data processedflow is designed and suited to hyperspectral engineering application.The innovations of this paper include the following three points.In the view of unminxing algorithm, a new spectral unmixing algorithm isproposed based on that the component of the unconstrained least squares solutionvector of the linear spectral mixing model can be defined as the distance from thepixel to the endmember. And a new classification method of endmember extractionalgorithm is proposed based on the different design mechanism. At the same time, aproperty is proposed and proved of linear spectral mixing model (the section of2.3.2of the paper).From the view of data processing and minernal tridimensional exploration, atechnical system of hyperspectral remote sensing used in resource exploration isproposed and designed.A data process flow based on airborne hyperspectral engineering application isdesigned, which includes the spectral reconstruction based on empirical linear,endmember extraction based on PPI, endmember identification based on the spretralproperty, and the mapping algorithm based on MTMF and threshod segmentationtechniques.