The Design And Implementation Of An Automatic Analysis Tool For Heavy Minerals In River Sand

The accurate and high efficient identification of heavy minerals has great signif-icance for source analysis,deposit prospecting and geological exploration,etc.The traditional manual identification requires higher professional knowledge,and has poor reproducibility and low identification efficiency.With the extension of geological in-struments,researchers analyzed system by electron microscope scanning and spec-trometer to obtain the major components and contents of heavy minerals,and also sought to tap the links between these energy spectrum data and heavy mineral classes,then to reach the purpose of heavy minerals identification.However,due to the large scale of energy spectrum data and lack of effective storage and management,which has affected the development of related research.In the light of the problems above mentioned,this paper uses rule learning tech-niques to design and implement an automatic analysis tool for heavy mineral that in-tegrates information storage,query,visualization,and rule mining.This tool provides users with information management and analysis for heavy minerals,and users can use rules mining algorithms to automatically extract the classification rules of heavy mineral,providing the basis for users to determine the category of heavy minerals.According to the needs of information storage and automatic analysis of heavy minerals,this paper designs and implements a automatic analysis tool for heavy min-erals.The main contributions of this paper include:(1)Summarize existing methods of rule miningThis paper first expounds the concept of rule learning,and analyses two rule min-ing frames which are top-down induction and sequential covering,introducing repre-sentative methods of the frames.(2)Propose a rule generation method which bases on radial basis function net-workThis paper proposes a rule generation method called Screnn which bases on radial basis function network.The method uses soft clustering in order to determine the memory centers of hidden neurons of the neural network.Next,Screnn extracts rules from trained neural network by using gradient descent with momentum,iterating and updating rule set until convergence constantly.Screnn not only reduces size of rule set,but also boosts the classification performance of rule set.(3)Analyse requirements of heavy minerals tool and design itThis paper uses case analysis in order to determine the functionalities of tool which is able to manage heavy minerals' data,inquiring information of heavy min-erals,visualizing spectrum data,and learn rules from spectrum data.Next,this paper divides the tool into database which is a basis module and functional modules such as information exhibition.Besides,this paper designs the tool from user interface and function of each module,and finally confirms the designing scheme of tool.(4)Explore validity of rule mining methods and designing scheme of toolThis paper investigates some research questions which are classification accuracy of rule set,quality of rule set and value of parameters.Based on spectrum dataset of heavy minerals,this paper designs and proceeds experiments,using some evaluation criteria to assess rule mining method such as Screnn.The results of experiments indi-cate that those rule mining methods can extract rules from spectrum data effectively.In addition,this paper implements modules from the point of user interface as well as function,aiming at the basis and functional modules of heavy minerals tool.By running the tool successfully,this paper verifies designing scheme of tool.