基于GF-2影像的离子吸附型稀土矿含量推断方法

Ion-adsorption rare earth ore content inference method based on GF-2 image

寻找离子吸附型稀土矿床对保障我国关键矿产资源具有重要作用。综合利用广西崇左六汤矿区的基础地质、地球化学勘探、高分辨率遥感影像等多源地学数据,以已知矿床分布特征为约束,基于多项式回归及BP神经网络对该区进行建模,以决定系数R 2及均方根误差(RMSE)为模型评价指标,对研究区离子吸附型稀土矿含量进行预测。研究结果表明,多项式回归模型检验R 2=0.54,BP神经网络模型检验R 2=0.64,剔除数据中高离群值后模型精度显著上升,多项式回归模型精度较好,但预测效果图与实测效果图差异较大。综上,数据中离群值的存在对模型的影响较大,模型拟合的好坏并非判断模型好坏的唯一标准,BP神经网络模型能较好预测研究区离子吸附型稀土矿含量。

The search for ion adsorption rare earth deposits plays an important role in safeguarding China′s key mineral resources.Taking the Liutang mining area of Chongzuo City,Guangxi Zhuang Autonomous Region as the research area,this paper comprehensively utilizes the basic geological data,geochemical exploration data,high-resolution remote sensing images and other multi-source geological data in the area.Polynomial regression and back propagation neural network(BP neural network)were used to model and invert the region.The distribution of rare earth ore content has been predicted from model evaluation indicators of the coefficient of determination R 2 and root mean square error(RMSE).The research results show that,firstly,the polynomial regression model test R 2 is 0.54,and the BP neural network model test R 2 is 0.64.Secondly,after removing the high outliers in the data,the model accuracy increases significantly.Thirdly,the polynomial regression model has better accuracy,but the predicted map is quite different from the measured map.In summary,the existence of outliers in the data has a great impact on the model,and the quality of the model fitting is not the only criterion for judging the quality of the model.The BP neural network model can better predict the content of ion-adsorption rare earth deposits in the study area.