Study On Methods Of Estimation And Monitoring Soil Heavy Metal Content In Mining Areas Based On Machine Learning Using Remote Sensing Technology

In view of the problem that the estimation accuracy of heavy metal content in the mining area based on multispectral remote sensing images is not high,this paper takes the Daxigou iron mining area in Shaanxi Province as the research area,and uses the spectral index,DEM extracted from the Landsat8 / OLI image spectral band,and physical and chemical analysis of the measured soil samples The data is the data source.By studying the correlation analysis between soil heavy metals and vegetation and terrain,a set of impact factors that can indirectly reflect the spatial distribution of soil heavy metals is constructed.A genetic algorithm(GA)optimized BP neural network model is established and selected 80% of the sample data trained the model to achieve the spatial estimation of the content of three heavy metals in the soil in the study area in 2017.Furthermore,for problem of high cost of training samples for multi-temporal soil heavy metal content monitoring tasks,based on the GA-BP network model established in 2017,the parameters of its GA-BP model were selectively transferred to the 2019 heavy soil metal estimation task in the study area,and a BP neural network model based on parameter transfer learning was innovatively constructed.The estimation of the heavy metal content of the soil in the study area in 2019 was achieved.The main findings and conclucsions are as follows:(1)The GA-BP network model was established by comprehensively using the spectral reflectance factor,spectral index factor and topographic factor of Landsat8 images,and the model was verified to estimate the content of three heavy metal elements in the soil in the 2017 study area.It is feasible,in which the estimation accuracy of copper and lead elements is relatively high.(2)Use the remaining 20% of the test data to compare and analyze the estimation errors of the multivariate linear model and the GA-BP neural network model.The results show that the GA-BP model established in this paper has three elements of copper,lead and arsenic in the 2017 study area The estimated root mean square error(RMSE)is much lower than the RMSE value of the multivariate linear model.This also shows that,compared with other commonly used methods,the GA-BP model has greatly improved the estimation accuracy of the three elements.(3)According to the spatial estimation results of the content of three heavy metals in the soil in 2017,the content of copper,lead,and arsenic in the soil in some areas of the study area exceeded the maximum value in the local national statistical data,while the content of copper and arsenic in the soil The values exceed the average of the background values,which indicates that the content of these three heavy metals in the soil of the area has a trend of increasing.And the three kinds of heavy metals are mainly distributed in the mining area,slag stacking area,both sides of the road and the bottom of the slope.This conclusion is consistent with the field survey results,and also verifies the use of the GA-BP neural network model to estimate the heavy metal content of the soil in the study area.Feasibility and effectiveness.(4)The GA-BP model based on parameter transfer learning can be used to selectively migrate the parameters of the GA-BP model in the 2017 source domain to the target domain in different phases of the same region,and then combine the fewer The soil sample data is used to estimate the heavy metal content of the soil in the target domain,and compared with the traditional neural network model that does not use transfer learning to verify its feasibility and reduce the cost of acquiring soil sample data.