基于XRF-CNN土壤重金属Zn元素含量预测模型研究  

Research on Prediction Model of Soil Heavy Metal Zn Content Based on XRF-CNN

作　　者：陈颖[1] 杨惠 肖春艳[2] 赵学亮 李康[3] 庞丽丽[3] 史彦新[3] 刘峥莹[1] 李少华 CHEN Ying;YANG Hui;XIAO Chun-yan;ZHAO Xue-liang;LI Kang;PANG Li-li;SHI Yan-xin;LIU Zheng-ying;LI Shao-hua(Hebei Province Key Laboratory of Test/Measurement Technology and Instrument,School of Electrical Engineering,Yanshan University,Qinhuangdao 066004,China;School of Resources and Environment,Henan University of Technology,Jiaozuo 454000,China;Center for Hydrogeology and Environmental Geology,China Geological Survey,Geological Environment Monitoring Engineering Technology Innovation Center of The Ministry of Natural Resources,Baoding 071051,China;Hebei Sailhero Environmental Protection High-tech Co.,Ltd.,Shijiazhuang 050035,China)

机构地区：[1]燕山大学电气工程学院河北省测试计量技术及仪器重点实验室,河北秦皇岛066004 [2]河南理工大学资源与环境学院,河南焦作454000 [3]中国地质调查局水文地质环境地质调查中心,自然资源部地质环境监测工程技术创新中心,河北保定071051 [4]河北先河环保科技股份有限公司,河北石家庄050035

出　　处：《光谱学与光谱分析》2021年第3期880-885,共6页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划项目(2018YFC1800903,2016YFC1400601-3);河北省重点研发计划项目(19273901D,20373301D);河北省自然科学基金项目(F2020203066);中国博士后基金项目(2018M630279);河北省博士后择优资助项目(D2018003028);河北省高等学校科学技术研究项目(ZD2018243)资助。

摘　　要：结合X射线荧光光谱法,针对土壤中重金属元素Zn含量的预测问题,提出基于深度卷积神经网络回归预测模型。对原始土壤进行相关预处理,用粉末压片法制作土壤压片,采用X射线荧光光谱法(X-Ray-fluorescence,XRF)获取土壤光谱,相比于传统检测方式,XR F法具有检测速度快、精度高、操作简单、不破坏样品属性并且可实现多种重金属元素同时检测等优点,故将XRF与深度卷积神经网络相结合,实现对土壤中重金属Zn元素含量的精确预测。采用箱型图来剔除X射线荧光光谱中的异常数据,采用熵权法结合多元散射校正来对样品盒数据进行校正,采用Savitzky-Golay平滑去噪法以及线性本底法对光谱数据进行预处理,可以有效地解决由外界环境和人为因素产生的噪声及基线漂移等问题。针对卷积神经网络结构的特殊性,将获取的一维光谱数据向量,采用构建光谱数据矩阵的方式来进行处理,将同一浓度、同一含水率下5组平行光谱数据向量转化为二维光谱信息矩阵,以该矩阵作为深度卷积神经网络预测模型的输入,以适应卷积层的操作要求,利用深度卷积神经网络特殊的结构模式,能有效提取土壤光谱数据特征,提高了深度卷积神经网络预测模型的学习能力,降低模型的训练难度。深度卷积神经网络预测模型采用3层卷积层搭建,使用ReLU激活函数激活,采用最大池化方式,减少数据的维度,增加Dropout层,防止过拟合,使用ADAM优化器对预测模型进行优化。实验以平均相对误差(mean rel ative error,MRE)、损失函数(LOSS)、平均绝对误差(mean absolute error,MAE)确定了模型的最优学习率为10^(-3)以及最优迭代次数为3000,并将深度卷积神经网络预测模型与BP预测模型、ELM预测模型、PLS预测模型进行对比,以均方误差(mean square error,MSE)、均方根误差(root mean square error,RMSE)、以及拟合系数R^(2)来分析比较预测模型的Combined with X-ray fluorescence spectroscopy,a prediction model based on deep convolutional neural network regression is proposed to predict the content of heavy metal element Zn in soil.Related pretreatment of the original soil,and soil compaction by powder compaction method,and the soil spectrum was obtained by X-Ray-fluorescence(XRF).Compared with traditional detection methods,the XRF method has the advantages of fast detection speed,high accuracy,simple operation,non-destructive sample properties,and simultaneous detection of multiple heavy metal elements.Therefore,XRF is combined with deep convolutional neural networks to achieve Precise prediction of heavy metal element Zn content in soil.In the experiment,box plots were used to eliminate abnormal data in the X-ray fluorescence spectrum.Entropy weight method and multiple scattering correction were used to correct the sample box data.The Savitzky-Golay smooth denoising method and linear background method are used to preprocess the spectral data,which can effectively solve the problems of noise and baseline drift caused by the external environment and human factors.The obtained one-dimensional spectral data vector is processed by constructing a spectral data matrix,this method converts 5 sets of parallel spectral data vectors at the same concentration and the same water content into a two-dimensional spectral information matrix and uses this matrix as the input of the deep convolutional neural network prediction model to meet the operational requirements of the convolutional layer.The learning ability of the deep convolutional neural network prediction model is improved,and the training difficulty of the model is reduced.The deep convolutional neural network prediction model is built with 3 layers of convolutional layers and activated using the RELU activation function.The maximum pooling method is used to reduce the dimensionality of the data and increase the Dropout layer to prevent overfitting.The ADAM optimizer is used to optimize the prediction model.Th

关 键 词：土壤重金属 X射线荧光光谱 光谱信息矩阵 深度卷积神经网络 

分 类 号：X833[环境科学与工程—环境工程]