Predicting TFe content and sorting iron ores from hyperspectral image by variational mode decomposition-based spectral feature

Abstract

During iron ore processing, accurately predicting the total iron (TFe) content in the ore is important for highly efficient and accurate sorting, which can reduce dust pollution, energy consumption, and economic losses, and promote waste utilization. Traditional techniques have limitations in accurately predicting the TFe content of individual iron ores, whereas hyperspectral imaging (HSI) has the potential to fulfill this task. Therefore, this paper explored it. For the experimental materials, 150 high-grade and 150 low-grade ore samples with particle sizes ranging from 20 to 40 mm were prepared. Subsequently, hyperspectral images of these samples were acquired within the range of 953–2517 nm, and the TFe content of each ore sample was measured. After preprocessing, the spectral features were extracted from the mean spectra of the iron ore, based on the analysis of the ore samples using variational mode decomposition (VMD). Subsequently, a random forest (RF) model based on these features was proposed to predict TFe content of the ores, and further sorted each individual piece of iron ore. Based on the evaluations, the prediction coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE) and classification accuracy of the proposed model, VMD Feature-random forest (VMD-RF), were 0.94, 0.07, 0.03, and 0.91, respectively. The results proved the feasibility of accurately sorting each individual iron ore based on its predicted TFe content using HSI. This methodology offers technical support for a cleaner iron ore production.