Given the limitation of conventional soil pollution monitoring through mapping which is a costly, time-consuming work, the study aims to establish an image recognition model to identify the source of pollution automatically. The study choses a contaminated land and then use a non-destructive instrument that can quickly and effectively measure the content of heavy metals. A two concentration prediction models of Ni, Cu, Zn, Cr, Pb, As, Cd, and Hg using hyperspectral imaging were developed, Decision Tree and Back Propagation Neural Network, in combination of particle swarm optimization employed for optimization algorithm. As a result, random forest is more accurate than the forecast result of back propagation neural network. This study has established an excellent Cu and Cr model, which can accurately capture the pollution source. In addition, through aerial photographs, we also found that there were also high pollution reactions on the banks of the river. The developed model is beneficial for high pollution areas which can be quickly found, thereby following investigation and remediation work can be carried out with less time and cost consuming comparing with the conventional soil monitoring.
Keywords: Heavy metals; Hyperspectral sensing; Neural network; Random forest; Soil pollution; UAVs.
© 2022. The Author(s), under exclusive licence to Springer Nature Switzerland AG.