Research On Lithology Extraction Method Of Limestone In Northeast Yunnan Based On ASTER Image

Geospatial information data acquisition technology,represented by satellite remote sensing,can quickly,accurately and extensively acquire surface information such as mineral resources distribution,road traffic status and land use on the earth's surface.How to use remote sensing technology to extract lithology more accurately on the basis of existing data is still a research hotspot.Based on a large number of field sampling data accumulated by Huize lead-zinc deposit research,this paper takes the widely distributed limestone lithology in the study area as the extraction target,and uses ASTER image as remote sensing data source for lithology extraction in the study area.ASTER image was launched in 1999,with a spatial resolution of 15 m and a high spectral resolution.Compared with Landsat-8,ASTER image has an advantage in spatial resolution.Excessive spatial resolution images such as SPOT will result in a large number of mixed pixels gathering in the interior of the object and easily lose the macroscopic characteristics of the geological body.The biggest advantage of ASTER image is that 14 bands can be set better for geological body recognition.The bands set near 2.3?m are more dense than Landsat-8 image.The hyperspectral curve of limestone can still retain absorption characteristics after resampling to ASTER band.ASTER image is more effective than Landsat-8 image in extracting lithologic distribution information of limestone and distinguishing fine differences in carbonate rocks.So ASTER is chosen as the image data source of this study.The pretreated ASTER images were extracted and analyzed by mineral index method,principal component analysis method,ASTER thermal infrared band and spectral angle method.The large difference of terrain height and high vegetation coverage in the study area cause great disturbance to the accuracy of lithology extraction.Using single lithology extraction can not make full use of the information in remote sensing images and eliminate the disturbance.It is still not objective in determining the threshold of abnormal values.A lithology extraction method based on decision tree is proposed.Before extracting lithology by mineral index method and principal component analysis method,disturbance information such as vegetation and water body is masked to improve the extraction accuracy of the two lithology extraction methods.The extraction results of mineral index method and principal component analysis method are superimposed with the information of Ca element abundances on the surface(based on the inversion formula of ASTER thermal infrared band and Ca content on the surface)and the extraction results of spectral angle(matched image pure end-element spectrum,obtained by comparing the overall similarity of spectrum),and the actual lithologic distribution of the study area is synthetically studied,and a CRT-based growth formula is established.The lithology extraction rules of the decision tree of the method.The lithology is extracted by decision tree rules,which can automatically identify lithology by computer,reduce human interference and improve recognition accuracy.Compared with the single lithology method,the threshold of lithology extraction by decision tree rule is obtained by machine learning,which is not affected by subjective factors.At the same time,the rule of decision tree is based on four kinds of lithology extraction results.It can extract lithology information synthetically by utilizing four characteristics of spectral information: characteristic absorption band,overall spectral shape,element content abundance and image occurrence information.After field inspection,the extraction accuracy reached 87.4%.The branch and leaf nodes in decision tree rules are screened by computer,and they are composed of various indicators which have greater benefits for lithologic classification.The construction of these nodes can be used for reference in other regions with similar geographic and climatic environment,such as northeastern Yunnan,when using decision tree to extract lithology.