Study On Hyperspectral Remote Sensing Methods For Rock Classification And Mineral Identification In Vegetation Covered Area

Rock classification and mineral identification by remote sensing is animportant tool for mineral investigation at a large scale. In vegetation covered area,the rock and mineral on remote sensing image are minor targets and weekinformation. The method on rock classification and mineral identification fromhyperspectral data in vegetation covered area are studied in this thesis, whichprovide technologic reference for geological prospecting.The study area is located in Human, Heilongjiang Province, where thevegetation covers as much as95%and there exists thick soil layer. The spectra ofvegetation, Layer A, B and C of soil, as well fresh and weathering rock, werecollected by ASD spectrograph in2010to2012at9sample points.According to geometric and geological features in study area, soil-vegetationindex and non-soil vegetation index are introduced after analyzing differentvegetation indexes. Different vegetation indexes after principle componenttransformation could separate vegetation from rock and soil. The rock-soilcomponent and vegetation component are analyzed by scattering plot after principlecomponent transformation of soil-vegetation index. The scattering plot couldseparate further soil and rock, which considers background as vegetationinformation and other integral information, and surrounding abnormal as soil-rockinformation. After the abnormal information is outlined, the average spectra arecalculated. Thus the rock classification is obtained by determining the abnormaltype from spectra analysis. The classification result is overlaid to the geological map.Some rock types are matched well with the geological map, including rhyolite,andesite tuff, glutenite, quartz,schist, marble, and basalt. Their Kappa coefficientcalculated by geological map and classification result is0.657and the overallaccuracy is71.8%, both of which satisfies the precision of rock classification.Based on the spectra of mineral, plants, and soil, the vegetation-depressionbased mineral identification method using hyperspectral data has been proposed. Itis then applied to EO-1Hyperion hyperspectral data. Six largest bands are selectedfor the difference calculus when the vegetation reflectivity is much different fromthe mineral reflectivity. The vegetation is depressed and the abnormal information is enhanced by analyzing bands after difference calculus. After vegetation depression,new data are generated for principle component transformation when the differenceof two principle components is the biggest difference among eigenvalues ofprinciple component. The abnormal information is outlined in the plot scattering.The spectral analysis has not been carried out until the average spectra arecalculated by abnormal information. In the area that some minerals are identifiedand some sample points are determined to validate the mineral identification results.It shows that some minerals are found in the sample points, including iron oxide,muscovite, biotite, pyritization and chlorite.On the basis of linear spectral mixture model, the spectra of mineral, dry grassand green plants are mixed by the contents increase of10%, and the spectralmixture is obtained. Calcite and kaolinite are taken as examples of mineral spectra.The spectral absorption position of calcite, kaolinite, green grass and dry grass, are2330nm,2200nm,670nm,2080nm, respectively. By analyzing absorption depths ofmixed spectra, there exists a negatively linear correlation between the absorptiondepths of calcite at2330nm and mixed spectra and that of green plant at670nm,wet grass at2080nm respectively. The linear relationship between mixed spectra andvegetation, dry grass, and mineral samplings, are determined by the theory of leastsquare method. It is applied to EO-1Hyperion image. Results show that, the fittingrelationship between mixed spectra and single sampling has a greatly significantlevel. Comparing the extracting mineral results and field results of rock, theextracting result is also good, but the disturbance information is found in the riverbed or along the road.