The Analyse And Abstraction Of Mine Spectra Feature On Hyperspctral Image

Presently, the method of hyperspectral remote sensing in mine analyzes can be reduced to two ways. One is the traditional statistic computation of the image data and bands, which typical includes supervised classification, maximum or minimum noise faction, spectral angle mapping etc. The other is the analyzes based physical elements of spectral feature, such as hydronium absorbing bands, spectral absorbing feature etc. Each has his strong point and weak point. The traditional statistic computation of the image data and bands can compute the relativity by the aspects of space, frequency and spectra to statistic the feature. But the method does not consider the physical chemistry processing of imaging that is hardly to coincide the mine element to spectra feature. The advantage of analyzes method based physical elements of spectral feature is it analysis the spectra, researches deeply the corresponding relatives among spectra absorbing feature, mine physical chemistry factor and geometry scaled. But the method is short of mathematics calculation. After the comparing of existing methods, it is better to band together both two methods to synthesis analysis the feature of spectra in order to make up the lacks. The article firstly collects and compares the algorithms of spectra at home and abroad, then find synthesis the imaging processing, mathematics calculation and physical chemistry meaning to researches and abstractes the feature of mine spectra becomes the trend.This paper firstly introduces and compares briefly the development of the hyperspectral remote sensing mine spectra feature at home and abroad, then summarizes the application of hyperspectral technolodge on geologic investigation. Secondly summarizes the preprocessing of hyperspectral images and aiming at the ordinary existents of mixed pixels in hyperstral image, advances the IEA algorithm to abstract the endmembers. Thirdly, integrates the materializes of the features of mine spectra and points out the method to calculate and analyze the feature of mine spectras. Lastly the author codes the algorithms and carry out it on AVIRIS data toexperiment and approve the algorithms' veracity.The kernel of paper surrounds three aspects to research. Firstly, aiming at the ordinary existents of mixed pixels in hyperstral image, advances the IEA algorithm to abstract the endmembers. Secondly, discusses the calculation and abstraction method of mine spectra from hyperspectral image, including feature factors abstraction on continuum lines, spectra relationship statistic, spectra index calculation and spectra's derivative and differential coefficient computation. Lastly, author codes the programs of the algorithm and tests it on hyperspectral data to validate the method. The mainly productions and conclusions are as follow:Aiming at the ordinary existents of mixed pixels in hyperstral image, the IEA algorithm can search the endmembers from image. The basic flow is as follow: firstly an initial vector (usually the mean spectrum of the data) is chosen to start the process. Then a constrained unmix is then performed and the error image is then formed. The average of the vectors with the largest error (distance from the initial vector) is assumed to be the first end-member. Another constrained unmix is then performed and the error image formed. The average of the vectors with the largest error (distance from the first end-member) is assumed to be the second end-member. Lastly this process is continued until the predetermined number of endmembers is found. The most advantage of IEA algorithm is the endmembera are generated spectras from the image itself, which means that unit conversion is not an issue, since the units will always agree, such as imaging sensor, wavelengths, band-widths, band shapes, times, Solar angle, terrain, climate etc. And the other benefit of it is avoids the happens of the endmembers are not existents actually. Secondly, summarizes the spectra's feature and shape of familiar hydroniums in mine, such as Mn²⁺, Fe²⁺, Fe³⁺, AL³⁺, OH⁺, CO₃²⁺ ect. And discusses the calculation and abstraction method of mine spectra from hyperspectral image, including feature factors abstraction on continuum lines, spectra relationship statistic, spectra index calculation and spectra's derivative and differential coefficient computation. Lastly synthesis the imaging processing, mathematics calculation and physical chemistry meaning to researches and abstractes the feature of mine spectra.Lastly, on the platform of MAPGIS, using VS2005 as the coding tool, codes the algorithms of mine spectra feature calculation. Then carries out and tests the AVIRIS hyperspectral image to analysis and proves the validity and practicalbility of the research.