Rocks Classification By Their Spectra In Geological Corridor Of Western Liaoning Province

Rocks are the information carrier in the evolution of the earth,recording the geological events occurring on earth during the evolution.The study of rocks will help us to better understand the evolution of the earth.At the same time,combining the study of rocks with the research in branches of geology such as geochemistry and geophysics,will help to solve major theoretical problems in geology.Although traditional methods of lithology information acquisition such as chemical composition analysis and physical properties measurement can obtain accurate and reliable information of rocks,the limitations in space and time still exist.Therefore,modern rock analysis requires more efficient,non-destructive and wide range of rock information extraction methods to facilitate the development of the industry.The research of remote sensing lithology information extraction has attracted more attention.This paper chooses the geological corridor in western Liaoning province as the study area,the area is developed with magmatic rocks and sedimentary stratum and is an ideal area for the study of lithological information extraction using remote sensing technology.On the basis of petrographic identification,measurement of indoor reflection spectra and the analysis of 48 chemical components of rock samples collected in the field,this paper found the relationship between the rocks chemical elements and the rocks reflectance using the single variable linear regression and multivariate linear regression statistical method.Modeling results show that the content of MgO,SiO₂,P₂O₅,Fe₂O₃ in the magmatic rock samples have a good correlation with the spectral reflectance information,and the content of Al₂O₃,CaO,P₂O₅,Ga in the sedimentary rock samples have a good correlation with the spectral reflectance information.And the PLSR model is the best model to predict the content of SiO₂ in the magmatic rock samples.Based on the continuum removed processing of spectral reflectance of rocks,principal component analysis is used to extract principal component components which accumulate more than 85%information,and classify the rock samples using Fuzzy c-means classification method.And all the rock samples can be divided into intrusive rock class,volcanic rock class and endogenous sedimentary rock class,the best classification accuracy can up to 82%,the overall classification accuracy is 64%.According to Hyperion and ASTER spectral response function of remote sensing image,resample the rocks ground spectral reflectance respectively,and aim at to get the model of predict the content of SiO₂ and the method of rocks classification in different kinds of remote sensing images.The PLSR prediction model for SiO₂content was established on the resampling band,which can predict the SiO₂ content of sample points well;Verified by randomly selected sample points from two kinds of image,using the rock classification method based on ground spectral model is better than the traditional supervised classification method.All the results verify the feasibility of rock classification method applied on the remote sensing images.