Evaluation And Application Of Mineral Resources Potential In Vegetation Coverage Area Using Remote Sensing And GIS

Mineral resources are the important material basis of social development,thus the development and utilization of mineral resources are inevitable requirement of modernization.However,with the progress of mineral exploration,the shallow minerals always could not meet the demand,and traditional geological prospecting methods become difficult to break through.At the same time,the development of new energy sources such as wind energy,solar energy and so on are in the initial stage,and the conversion efficiency is still relatively low.Therefore,the comprehensive utilization of geological information and geophysics,geochemistry and remote sensing data,using computer data mining method to realize mathematical statistical prediction models,combined with the actual characteristics of the study area,to evaluate the regional mineral resource potential effectively and establish comprehensive information mineral resources prediction model then delineate target areas of high probability,is the main research direction at present.It has important practical significance in reducing search cost and improving accuracy.For the demand of mineral resources potential evaluation in industry department,three methods of potential assessment in mineral resources are provided in this thesis.Especially solve the problem that it is relatively difficult to find mineral resources in higher vegetation covered area.Finally,the thesis takes Changbai area in Sichuan province as example to conduct detailed apply.The main work and achievements are as follows:(1)Take account of the situation that the research area has higher vegetation coverage in most cases,we use ASTER data combined with a series of algorithms to extract vegetation information,including the use of the minimum noise fraction(MNF)technology to extract relatively pure bands,the sequential maximum angle convex cone(SMACC)at pixel scale to gain the abundance in each end member,spectra feature fitting(SFF)classification and spectral angle Mapping(SAM)technique to identify vegetation end member,finally using linear decomposition compensation replacement method to gain weaken the effect of vegetation remote sensing data in a certain extent.Last but not least,take advantage of the reflection peak and absorption valley in the target mineral specific spectral curve,the remote sensing alteration anomaly is extracted by the band ratio method.(2)Build deposit model according to existing prior data and actual situation,and combine the decision tree theory to determine the geological,geochemistry and remote sensing attributes which have intimate connection with mineralization.And then make all the attributes in same consistency processing with the help of geographic information system(GIS).(3)Establish mineral resource potential evaluation models based on evidential belief functions and Logistic regression model.The former is through the basic probability assignment function to define the belief function,plausibility function,support function and uncertain function as evaluation index;the latter gain the favorable ore probability by maximum likelihood estimation,and then using the Newton iteration to find the optimal solution.(4)Establish a mineral resource potential evaluation model based on the method of restricted Boltzmann machine(RBM).The application of RBM solves the problems of mineralization information shortage and lack of known sample points which are often encountered in the actual situation of mineral resources potential evaluation.It uses the idea of deep learning to establish the objective function,and carries out the contrastive divergence algorithm(CD algorithm)for all the targets.Thus we can gain the target area without any prior knowledge.(5)Using statistical method,ROC curve and AUC value to evaluate the accuracy of the model.Through the evaluation method in the horizontal and vertical comparison results,we can conclude that weaken the effect of vegetation influence could obtain more accurate results than original data;based on RBM model of deep learning is more effective than the traditional statistical model.