Mineral Mapping Based On Hyperspectral Image Feature Space

With the development of science and technology,the applications of remote sensing technology in mineral resources exploration are more and more extensive.Hyperspectral remote sensing technology,which can accurately extract mineral alteration information,has become one of the key means to obtain mineralization information in the early stage of mineral exploration.Hyperspectral image data contain rich spectral information,but there are many bands,most of which have high correlation,and there is obvious information redundancy phenomenon.Moreover,some bands of hyperspectral image are seriously affected by noise,which is not conducive to correctly identify and map rock and minerals.After the hyperspectral data is converted in the principal component feature space by using the dimensionality reduction method,the number of bands can be effectively reduced,most of the noise can be eliminated,eliminate vast majority of noise,and eliminate data redundancy.The generated feature principal components only contains the feature information of the original image,which is very suitable for subsequent geological information extraction.Therefore,based on the hyperspectral characteristic spatial data,this paper studies the hyperspectral image noise removal and mineral mapping,designs an efficient,high-precision and robust hyperspectral image de-noising and mineral mapping framework,and discusses the application effect of this mineral mapping framework in AVIRIS airborne hyperspectral data and GF-5 spaceborne hyperspectral data.The main research contents and results of this paper are as follows:(1)Research on hyperspectral image noise removal.After conventional preprocessing,such as radiometric calibration,bad line repair,atmospheric correction and orthographic correction,hyperspectral images still contain significant noise in the feature space.In order to eliminate these noises effectively,the stripe noise removal technology and the salt-and-pepper noise removal technology based on the feature space of hyperspectral image are proposed in this study.Removal techniques,the new strip after pretreatment of hyperspectral images to remove the vegetation,water,snow,cloud,etc features,then pca transform,after selecting high order principal component for normalized per column of radiation,effectively remove sensor unit caused by bar belt,finally the principal component image transformation to the original image space,after effectively remove stripe images.In the new pepper and salt noise removal technology,the image after removing the stripe is firstly transformed by principal components,and then the high-order principal components are selected to carry out the Gaussian low-pass filtering scene by scene to effectively remove the pepper and salt noise in the principal components.Finally,the principal components are reverse-transformed to the original image space to obtain the image of effective pepper and salt noise.Through comparative analysis,it is found that these two techniques have significant effects on the hyperspectral image of Gaofen-5,and the quality of the processed hyperspectral image of Gaofen-5 has been significantly improved.(2)Hyperspectral image mineral mapping.In order to reduce the interference caused by redundant information and noise information in image features on mineral extraction,a new mapping framework of minerals based on hyperspectral feature space was designed in this paper.Reference this mapping framework,respectively with full band and mineral absorption peak band has designed two mineral mapping method.These methods were compared and verified in Cuprite mining area and Tiegelong mining area in Duolong.The results show that the design of the two new methods can effectively extract the main minerals in the study area,and has better robustness than the traditional technology.And the second method achieves the highest mapping accuracy.(3)The absorption peak bands of the main minerals are selected to map the minerals in the Duolong ore collection area based on the hyperspectral feature space mapping framework.The mapping results show that the alteration reflection in the Duolong ore collection area is relatively strong,and the distribution of minerals is consistent with the existing geological data.