Research On Extraction Of Remote Sensing Rock Information Based On Linear Mixing Spectral Model

Spectral unmixinghas been one of the hot issues in remote sensing research.Linear mixing spectral model is an effective method to solve the problem of spectral unmixing.According to the different constraints of unmixing algorithm,the spectral unmixing can be divided into unconstrained spectral unmixing,partiallyconstrained spectral unmixing and completelyconstrained spectral unmixing.There are many studies on the classification of land use,the extraction of crop area and vegetation cover,but the studies on lithology extraction is less and most of the lithology extraction is based on pixel-by-pixel.Therefore,this paper uses Hyperion spectral data to carry out the lithology extraction based on linear spectral unmixing at the junction between Dazi County and Muzhu Gongka County in Tibet as the research area.The main findings are as follows:(1)Data preprocessing of the Hyperion spectral data of EO-1 satellite.The real reflection image of the study area is obtained by data processing such as band rejection,stripes and bad line repair,atmospheric correction and image cutting.(2)Acquisition and analysis of rock spectraldata.By designing the rock collection routes and using the ASD spectrometer to obtain the rock spectral data in the room,the spectral data are processed and the lithology spectradatabase of the study area is established.On the basis of rock spectral data processing,the lithology spectral analysis is carried out to find out that the reflectivity of weathered surface is different from that of the fresh surface in the same lithology,but the characteristic of spectrum curveis basically the same.(3)By analyzing a variety of endmember extraction algorithms,an endmember extraction algorithm suitable for lithologic classification is elected.PPI,SMACC and N-FINDR are widely used and easy to implement.The three methods are used to extract the endmember of the study area,and the traditional SAM method is used to classify the lithology.The endmember spectrum and lithologic classification effect withing the three methods are compared and analyzed.The results show that the N-FINDR method is superior to the PPI method and the SMACC method.The PPI method due to human factors to make the extraction of the end element information is not complete and the endmember accuracy extracted by the SMACC method is relatively unstable.(4)Three abundance inversion algorithms of linear mixed spectral model are analyzed,and the optimal algorithm is selected for lithology extraction.By discussing the principle of linear spectral unmixing algorithm in three cases: unconstrainted least squares(UCLS),sum to one constrained least squares(SCLS),fully constrainted least squares(FCLS),and using three algorithms to carry out the abundance inversion of the study area.The RMSE values of the three inversion results are 0.0435,0.0378 and 0.0223,respectively.The overall accuracy is 87.71% and the Kappa is 79.82% by analyzing the lithology extraction of the algorithm of linear spectral unmixing.The results show that the accuracyof fully constrained least square is the highest,and the linear spectral unmixing based on the fullyconstraint can further improve the lithologic extraction accuracy compared with the traditional SAM lithologic extraction.