Remote Sensing Geochemical Elements Combination Inversion Model Based On Kernel Extreme Learning Machine

Geochemical exploration is an important means to obtain prospecting information in geological mineral exploration.Extraction of geochemical anomaly information is the key process to guarantee the effectiveness of geochemical exploration.However,geochemical exploration research inevitably involve a large amount of sampling work.It is extremely difficult to obtain large-scale geochemical data in inaccessible terrain with harsh working environment.With deepening of research on metallogenic theory,comprehensive analysis and prediction with integration of geochemistry and remote sensing technology have become the development tendency in ore prospecting.In this paper,a remote sensing geochemical inversion model is proposed which can deal with complex geological conditions and acquire large-scale and high-precision geochemical data in an efficient way,thus the ability of extracting geochemical anomalies can be improved.The key of remote sensing geochemical inversion is to establish the relationship between geochemical elements and remote sensing images.Existing inversion models do not consider their associated characteristics of geochemical elements under the complex geological factors.Factor analysis is an effective method that can dealing with elements association.Meanwhile,the distribution of geochemical anomalies has non-linear characteristics,e.g.,inconsistency,mutagenicity and diversity.It is difficult to obtain satisfactory results based on the traditional linear regression method,which directly affects the reliability of inversion.Kernel extreme learning machine(KELM)replaces the unknown hidden layer feature map with a kernel function.It has little computation requirement and strong generalization ability.Facing the geochemical anomalies extraction,this paper establishes elements combination with factor analysis to reflect geochemical element associated characteristics;then a nonlinear inversion model of remote sensing geochemical element combination based on kernel extreme learning machine is proposed.In this paper,the Sandaoqiao area of the Ordos Basin is used as the research area.The 1: 200,000 soil geochemical data and Landsat 8 OLI remote sensing images are utilized for inversion analysis.For important indicator elements,the geochemical elements combination is generated with factor analysis.For realizing the prediction and inversion of large-scale and high-precision geochemical data in the research area,the non-linear relationship between the geochemical elements combination and OLI images is established based on KELM.In model verification respect,a comparative analysis is made between single element and elements combination with extreme learning machines(ELM)and KELM.By analyzing the inversion prediction results,proposed inversion model obtained higher prediction accuracy and lower prediction error and show superior prediction ability.Meanwhile,known ore spots are selected to analyze the correspondence of abnormal regions.The experimental results show that the known ore spots is located in or at the edge of abnormal area by the inversion data obtained with proposed inversion model.Compared with the original geochemical data and comparison methods,it shows a better correspondence with known ore spots,which verifies the effectiveness of proposed model.The prediction data obtained by the proposed inversion model effectively supplement original geochemical data,and can be integrated with original geochemical data to extract the abnormal regions that cannot be determined using original geochemical data.Therefore,the proposed model can indicate ore prospecting direction and reduce the cost of geochemical exploration in areas with harsh environments.