Estimation of heavy metal concentrations in rice using support vector machines and particle swarm optimization from reflectance spectroscopy

Abstract

Heavy metal contamination of soils arouses serious problems in agriculture and poses a potential threat to the safe crop production. In this study，we explored the possibility of using the canopy reflectance of heavy metal stressed rice to estimate the concentrations of heavy metals, cadmium (Cd), arsenic (As), chromium (Cr), copper (Cu), lead (Pb), and zinc (Zn) in rice. Canopy spectral measurements from rice were collected using Analytical Spectral Devices (ASD) field spectrometer in situ in the crop lands of Suzhou, Jiangsu. Soil samples and rice samples were collected for chemical analysis of heavy metal concentrations. The normalization spectral pre-processing method was employed to improve the robustness and performance of the predicting model. Estimation of Heavy metal concentrations was achieved by a support vector machines approach. Particle swarm optimization (PSO) was applied for parameter optimization of support vector machines (SVM). The results show that the PSO-SVM approach yielded R2 values between estimated and measured heavy metal concentrations ranging from 0.40–0.89. The best estimation accuracies were obtained for Cd, Pb, and Cr. The poorest estimation was for Cu and Zn. The results indicate that it is possible to estimate concentrations of heavy metal in heavy metal stressed rice using reflectance spectroscopy and support vector machines.