Prediction Of Soil Copper Content Around Mining Area Based On Full Spectrum(VIS-NIR)and GF-5 Data

With the development and maturity of hyperspectral remote sensing,hyperspectral technology has been gradually used in various fields such as land and mineral resources exploration,and achieved good research results.Due to the copper mining in Dexing,Jiangxi Province,the soil copper pollution is serious.In order to quickly and effectively investigate the soil environment,the method of hyperspectral remote sensing is used to invert the copper content in the soil around Dexing copper mine in Jiangxi Province.Soil around in dexing copper mine area as the study area,combining with the copper content in soil samples data,soil samples indoor PSR ground spectral data and GF-5 remote sensing data,using machine learning method to establish ground spectral model with copper content in soil,and using the optimal model to invert the soil copper content from the remote sensing images in the study area,Finally,the copper pollution in the study area was evaluated.By analyzing the mechanism of soil spectra,the general laws of soil moisture characteristic bands and organic matter characteristic bands were found,and the relationship between soil particle size and tailings distance and soil spectral reflectance was found to be certain.The spectral law of organic matter,clay minerals and iron oxides in soil was analyzed,and the occurrence relationship of these three substances with soil copper was found,so the characteristic bands of soil copper could be inferred.The source and properties of copper in soil are analyzed.It is found that copper has the characteristics of migration,easy to combine with organic matter,solubility,easy to adsorb on clay surface,and can also form sulfide.The ground spectral data were transformed by first order differential,reciprocal logarithm,continuum removal and multiple scattering correction.By analyzing the correlation between the four transform spectra and the non-transform spectra and the copper content,it was found that the correlation value of continuum removal transformation was the highest,and the correlation bands of multiple scattering correction and first order differential were more.The partial least squares(PLS),BP neural network(BPNN)and support vector machine(SVM)models were established for the five spectra and copper content respectively.It was found that the optimal modeling R~2and verification R~2 of the PLS method were about 0.45,and the verified RPIQ was 1.628.The optimal modeling R~2 and verification R~2 of the BPNN method were more than 0.6.The verified RPIQ is 1.959,the optimal modeling R~2 and verification R~2 of the SVM method are 0.422 and0.4224,respectively,and the verified RPIQ is 1.6763.Comparing the accuracy of the model established by the three methods,the results of the BPNN model are better.The optimal results of the BPNN model are used to carry out GF-5 inversion,and at the same time,the inverse distance weight interpolation graph is made for the sampling points.The distribution of copper in the soil between the two is consistent,the closer to the mining area,the more copper in the soil.According to the copper content classification display,the copper pollution status was evaluated,and it was found that the copper pollution near the tailings dam was more serious,exceeding the risk screening value,and the pollution in other places was lighter.This study provides a scientific basis for local soil environmental governance,which reflects the importance of hyperspectral remote sensing soil survey and research.