Research on an XRF-visNIR soil heavy metal exceedance analysis method based on GAS transformation and PCANet

Abstract

Analyzing and rapidly screening the phenomenon of soil heavy metal exceedance remains a challenge for the fusion technology of X-ray fluorescence (XRF) and visible near infrared spectroscopy (visNIR). To address this, a new XRF-visNIR fusion method based on Gramian Angular Summation (GAS) transformation and Principal Component Analysis Network (PCANet) feature extraction is proposed. This method transforms XRF and visNIR data into two-dimensional images through GAS conversion, followed by feature extraction using PCANet. This reduces data dimensions and extracts important information about soil heavy metals. In the experimental phase, a large number of soil samples were collected from the Hongfeng Lake area and tested for spectral information using XRF-visNIR. By constructing and training a deep learning network, soil heavy metal pollution was classified and assessed. The results show that this method has achieved significant results in the analysis of soil heavy metal exceedance. The optimized GASF_PCANet_CNN can rapidly and accurately identify seven kinds of heavy metal pollution exceedance (Pb, Cd, As, Cr, Cu, Zn, and Ni). Deployed on an embedded platform, it can achieve quick feedback of screening results, with an average accuracy, average recall rate, average precision, and average F₁ score of 95.07%, 95.90%, 95.17%, and 95.53%, respectively. The XRF-visNIR soil heavy metal analysis method based on GAS transformation and PCANet proposed in this study provides an efficient and reliable analytical means for monitoring soil heavy metal exceedance, actively promoting soil pollution management and environmental protection work.