基于土壤背景估计算法和SpectraNet联合的XRF土壤重金属Zn超标分析方法研究

Research on the Analysis Method of Excessive Zn in Soil Based on Soil Background Estimation Algorithm and SpectraNet Combined with XRF

摘要: 随着社会现代化进程的迈进，频繁的人类活动加剧了土壤的重金属污染。当土壤中的重金属元素含量超过风险筛选值，会经过食物链摄入人体，过量的重金属累积对人体健康造成损害。因此，识别含有重金属污染风险的土壤样本，是开展污染防治工作的关键步骤。本研究采用X射线荧光(XRF)光谱仪获取了22份国家标准土壤样品的光谱数据，并结合airPLS与高斯卷积方法对光谱进行本底扣除等预处理操作；最后，利用本研究提出的深度学习模型SpectraNet对土壤样本是否存在重金属污染风险进行预测。实验结果显示，该模型在多个土壤元素数据集上展现出较好的识别性能和稳定性，本研究所提出的网络在多种土壤的定性分析任务中展现了更为全面的判别性能，在资源受限的情况下能够实现快速推理。

Abstract: With the advancement of social modernization, frequent human activities have intensified the heavy metal pollution of soil. When the content of heavy metal elements in the soil exceeds the risk screening value, they will be ingested into the human body through the food chain, and excessive accumulation of heavy metals can cause damage to human health. Screening out the soil with heavy metal pollution is an important step in soil pollution control. The spectral data of 22 national standard soil samples were obtained by using X-ray fluorescence (XRF) spectrometer, and then the background subtraction was carried out by combining airPLS and Gaussian convolution for pretreatment. Finally, the SpectraNet algorithm was used to predict whether the soil samples had the risk of heavy metal pollution. The experimental results show that the model has good recognition performance and stability on multiple soil element datasets. The network proposed in this study has shown more comprehensive discrimination performance in the qualitative analysis tasks of various soils and can achieve rapid inference under resource-limited conditions.