BACK-PROPAGATION NEURAL NETWORK–BASED MODELLING FOR SOIL HEAVY METAL

Fang Li,∗,∗∗ Anxiang Lu,∗∗ Jihua Wang,∗∗ and Tianyan You∗

Keywords

Levenberg–Marquart, back-propagation neural network, soil, heavymetal

Abstract

X-ray fluorescence (XRF) technology is a widely used method for rapid detection of heavy metals in soil. It is important to establish accurate models of the XRF spectrometer. Firstly, the influence of sample particle size on the detection results was investigated, and the 100 mesh was the most suitable particle size. All spectra were pre-processed before modelling using the Savitzky–Golay smoothing with seven points and the wavelet transform method with Coif 3 wavelet base at the ninth level for noise deduction and baseline correction. The error back-propagation artificial neural network (BP-ANN) learning algorithm optimized by Levenberg–Marquart (LM) algorithm was selected to establish soil heavy metal models. The results indicated that the modelling results were sufficiently accurate. The BP-ANN method optimized with LM algorithm was applied in the XRF research field for the first time, which provides some technical support for the establishment of rapid detection models of heavy metals in soil.

Important Links:

Go Back