Object-oriented Classification And Application Of Rare Earth Mine Land Occupation

Rare earth mines are important minerals and strategic resources in China.As rare earth mines have been mined in large quantities in recent years,the reserves gradually decrease,so it is urgent to accurately reflect the mining status of rare earth mines.The utilization of remote sensing technology in mine detection has unparalleled advantages.However,few authors have studied the land occupation method and dynamic monitoring of rare earth mines.The object-oriented method has a large excellences in results,but determining the optimal segmentation scale of features has always been a problem to be further studied.In this paper,the GF1 and ZY3 satellite images are used as data sources to carry out quantitative analysis of the optimal scale for the land occupation method of rare earth mines.On this basis,the nearest neighbor classification method is utilized for classification research,and the classification results are supported with support vector machine supervised classification to choose the best classification method.According to the classification results in 2015 and 2018,remote sensing dynamic monitoring of rare earth mines was carried out.The main conclusions:(1)An improved optimal segmentation scale evaluation method is proposed:the mean standard deviation is used as an index to gauge heterogeneity,the information entropy is used to gauge homogeneity index,an assessment means is constructed,and the variation curve means is integrated with visual analysis Obtain the optimal division scale of the rare earth mine land occupation method,and obtain the optimal division scale of the mining field,the beneficiation pool,the dumping field,and the mine buildings,respectively:60,80,60,and 100.(2)Combined with the features of remote sensing graphics of land occupation methods of rare earth mines,the features of spectrum,shape and texture are analyzed and extracted.Spectral features:spectral reflectance;shape features:area,extensibility,rectangular similarity.;texture features:entropy.,constructing classification rules,using nearest neighbor classification,the classification results is 88.89%,and the Kappa coefficient is 0.85.(3)Using support vector machine supervised classification means to compare with object-oriented classification,the support vector machine supervised classification accuracy is 81.42%,and the Kappa coefficient is 0.75.Through the qualitative analysis of the classification result graphs and the quantitative comparison of the classification accuracy,the accuracy of the object-oriented classification is higher and the classification results are better.(4)Dynamic monitoring and analysis of rare earth mines were carried out using the method of first classification and comparison,and it was obtained that the total area of rare earth mines mined in the study area from 2015 to 2018 was reduced by 9.16km~2,which was mainly restored and governed under the influence of relevant policies.According to the conclusions of moving detectation,relevant suggestions were made for the mining and management of rare earth mines in the study area.