Estimation Of Iron Concentration In Soil Of A Mining Area From Uav-Based Hyperspectral Imagery

The detection and monitoring of the iron (Fe) concentration for a mine area are necessary for environmental assessment. Traditional grid sampling plus interpolation approach or the advanced point-based spectroscopy via remote sensing can not effectively monitor Fe concentration in an area. Capturing hyperspectral imagery using unmanned aerial vehicles (UAVs) is a more efficient approach to retrieve and map the Fe concentrations for an Fe mine. In this work, we propose a pipeline of Fe concentration retrieval from UAV-based hyperspectral imagery using regression methods. Regression models considered include partial least squares (PLSR), support vector machine (SVR), and an artificial neural network (ANN). Different feature sets were also considered, namely, selected bands (SB), principal components analysis (PCA) reduction, and minimum noise fraction (MNF). Experiments demonstrate the strong potential for using UAV-based hyperspectral imaging for Fe monitoring. Model selection and feature extraction can greatly improve the retrieval accuracy of Fe concentration. A preferred approach that generated this highest regression success involved PLSR model trained with the combination of SB, PCA an MNF.