Research On Detection And Analysis Method Of Soil Heavy Metal Elements Based On XRF

With the continuous development of our industry and agriculture,soil pollution has become more and more prominent,among which soil heavy metal pollution is one of the most extensive and harmful problems.Therefore,it is of great significance to study efficient and accurate detection and analysis methods of soil heavy metals for understanding soil pollution and ensuring soil environmental safety.X-ray fluorescence spectroscopy has the advantages of simple operation,convenient and rapid,low cost and high precision,and has been widely used in the detection and analysis of heavy metal elements.However,in the actual detection and analysis,the original spectrum has high frequency noise and low frequency background,resulting in low correlation between spectral data and elements.At the same time,due to complex soil components and low heavy metal content,soil matrix effect and spectral line overlap and other factors,the relationship between spectral line strength and content of elements to be measured is nonlinear,which reduces the accuracy of quantitative analysis of heavy metal elements.Therefore,research on detection and analysis methods of heavy metal elements in soil based on XRF was carried out.The main work contents are as follows:(1)X-ray fluorescence spectrometer was used to detect the national standard soil samples,and XRF spectral images were obtained.The simulated spectra were generated by Monte Carlo simulation and normalized according to the characteristic peaks in different energy segments.The conversion between spectral channel and energy is realized by energy scale fitting.(2)A spectral data preprocessing algorithm based on improved wavelet packet denoising and background subtraction of generative adversarial network is proposed.The improved wavelet packet denoising method can correct the aliasing of spectral data band,and reconstruct the decomposed soil spectral signals from the perspective of energy distribution,which can filter out the low-frequency signals without losing the highfrequency data information.In addition,the generated adversarial network is combined with XRF spectra to realize the background deduction of spectra and obtain the corrected spectra.The proposed method is effective in spectral denoising and background deduction.The determination coefficients of the calibration curves of Cr,Ni,Cu,Zn,As and Pb elements were increased from 0.966,0.922,0.973,0.961,0.831,0.915 to 0.975,0.958,0.986,0.980,0.876,0.964.(3)Aiming at the problems of serious spectrum overlapping interference and matrix effect between elements in the analysis of soil samples by XRF,the competitive adaptive reweighting algorithm was used to select characteristic variables.On this basis,a fusion model of Harris Eagle optimized convolutional neural network and support vector machine was proposed for soil heavy metal detection and analysis.In this model,multilayer convolution is used to extract and fuse spectral feature information,and Harris Eagle optimization algorithm is used to find the optimal SVR hyperparameter to fit the relationship between heavy metal content and characteristic peak intensity.The model can simultaneously complete the detection of whether the heavy metal content in soil exceeds the standard and the quantitative analysis of heavy metal content,and solve the problem of low prediction results under small sample sets.The results showed that the prediction accuracy of heavy metal elements increased by 9.73% compared with the PLSR model.The coefficient of determination of training set was above 0.99,and the coefficient of determination of test set was above 0.89...