Using complex permittivity and artificial neural networks to identify and classify copper, zinc, and lead contamination in soil

Abstract

Complex permittivity measurements combined with artificial neural networks (ANNs) are investigated as a method for assessing and identifying heavy metal contamination in soil. The measurements are carried out with a custom-built device on 164 compacted samples of a natural clayey soil, artificially contaminated with various simple salts including heavy metals (Cu, Zn, and Pb). The soil samples are prepared by mixing solutions of the various salts with the soil at various concentrations and water contents. A database has been set up consisting of complex permittivity measurements made between the frequencies of 200 and 500 MHz and measured physical and chemical properties of the soil samples. Using this database as input, two ANN models are designed, the first to detect the presence or absence of heavy metals in the soil samples and the second to determine whether the heavy metal, if present in a given sample, is Cu, Zn, or Pb. Both ANN models perform reasonably well. Overall, the first model is able to detect the presence of heavy metals in 92.7% of cases, and the second is successful in distinguishing the particular type of heavy metal in 76.4% of all the samples containing heavy metals. These encouraging results underscore the potential of complex permittivity and ANNs as promising tools for nondestructive subsurface contamination assessment.