Studying Of Quantitative Retrieval Of Fe₂O₃and SiO₂Content In Surface Rocks

With the rapid development of national economy of our country, the demand formineral resources is rapidly increasing, and utilizing remote sensing in prospectinghas become a powerful method, so new requirements have been raised in prospectingthrough the extraction of remote sensing geological anomaly information.Hyperspectral remote sensing is more and more applied in fields such aspetromineralogy because of its high resolution. Besides hyperspectral remote sensing,thermal infrared band of ASTER data is also applied in this study, because it isdifficult for the hyperspectral visible light-near infrared band spectrums to detect thevibrational spectrums of Si-O of silicate minerals. Although Fe₂O₃contents in rocksare low, they could help determine the distribution of minerals such as limonite andhematite in rocks. These minerals are often closely associated with gold and metallicsulfide deposits. The content of SiO₂in rock is an important indicator in lithologyclassification, and quartz vein is also one of the main indicators of prospecting. Theseindicators together can provide direct or indirect information for us to look for theabove ore deposits.The modeling data used in this paper is based on standard digital spectrallibraries provided by United States geological survey （USGS）, the jet propulsionlaboratory （JPL）, and Johns Hopkins university （JHU）. This paper chose the sampledata through integrating these three spectral libraries, and used multiple stepwiseregression and partial least squares to establish13hyperspectral inversion models ofFe₂O₃contents, based on the original band reflectance, the high frequency part andlow frequency part acquired from the wavelet packet decomposition, and the highfrequency part and low frequency part acquired from the wavelet packetdecomposition after the continuum removal, respectively. There is a good correlationbetween the SiO₂contents in rocks on the earth’s surface and the rock spectrums, andusing multiple stepwise regression and partial least squares to model can acquire goodeffects. The main results and conclusions are as follows: 1. In the rocks on the earth’s surface, the relation between the Fe₂O₃contents andthe rock spectrums is complex, and has some uncertainty. This paper establishedmany hyperspectral inversion models of Fe₂O₃contents, based on the original bandreflectance, the high frequency part and low frequency part acquired from the waveletpacket decomposition, and the high frequency part and low frequency part acquiredfrom the wavelet packet decomposition after the continuum removal, respectively.The modeling methods are multiple stepwise regression and partial least squares. Theresults suggest, after the continuum removal, the spectrums have better effects interms of modeling, compared with the original spectrums. Models established byspectrums processed by wavelet packet analysis have better predictive ability andhigher precision, compared with the inversion models of the spectrums that have notbeen processed by wavelet packet analysis. The effects of high-frequency signals arebetter than that of low-frequency signals.2. In the rocks on the earth’s surface, there is a good correlation between theSiO₂contents and the rock spectrums. Using the thermal infrared bands of Aster datato conduct the inversion for SiO₂contents is feasible. The study shows, the ratios ofthe emissivities of the thermal infrared bands and the conversion of them have a goodlinear relation with the SiO₂contents. Using either multiple stepwise regression orpartial least squares to model can acquire good effects, and provide parameters andbasis for distinguishing one lithology from the others.3. After the continuum removal, the precision of modeling with multiplestepwise regression method and partial least squares method is improved, but theeffect of the inversion is unsatisfying. The theory of Xu-Yuanjin is confirmed: thecontinuum removal helps highlight the ground object spectral signature information,especially for vegetations whose waveforms have big fluctuations, however, forground objects whose spectral waveforms have small fluctuations, the information isnot obvious. The analysis of the results of identifying remote sensing images after thecontinuum removal shows, when the information of ground subjects such as rocks isenhanced, the background noise of the images will also be enhanced, resulting inunsatisfying identifying effects. Therefore, for hyperspectral images that have highbackground noise levels, adopting continuum removal and normalization processingare not in favor of the extraction of information of rocks and minerals.While modeling for the Fe₂O₃contents, it shows that the correlation coefficientsof the Fe₂O₃contents and the reflectances are generally low, and the SiO₂contentsacquired from the inversion are also lower than the actual values. According to the analysis, the reasons are as follows:1. Essentially, rock spectrums are mixed minerals’ spectrums, but their spectralsignatures are also affected by textures of rocks, structures and conditions of thesurface; even though the features of the absorption bands are the most effectivediagnosis parameters, these absorption bands became complex because they wereaffected by environmental factors such as vegetation, and their shapes and locationsmay change. Therefore, there are some uncertainties about inverting Fe₂O₃and SiO₂inrocks according to remote sensing images alone.2. The samples used in modeling are all from the spectral libraries, but thesamples in the spectral libraries can not represent the actual situations of the researchareas, such as the mineral associations of rocks and the exposure situations of rocks inthe research areas; besides, the atmospheric environments and the light conditionsacquired during the imaging of the remote sensing images are different from thatacquired during the indoor spectral analysis, and even if the correspondingatmospheric correction is conducted, the affection of these factors still can not becompletely eliminated.