Detection Of Heavy Metals In Chinese Cabbage And Its Growing Soil Based On Terahertz Spectroscopy

With the advancement of industrialization and the enrichment of material life,heavy metal pollution in vegetables and soil has attracted much attention.There are numerous detection methods for heavy metals,but most of the detection methods are complicated in the early stage of experiment.Based on Terahertz(THz)spectroscopy,qualitative and quantitative analysis of Hg,Cd and Cu in Chinese cabbage and its growing soil was carried out in this paper,aiming to explore the feasibility of THz spectroscopy for the detection of heavy metals in vegetables and soil,and to provide a simple,efficient and accurate new method for the detection of heavy metals.The research contents mainly include:(1)The frequency domain spectrum,refractive index spectrum and absorption coefficient spectrum of cabbage leaf and soil samples were analyzed based on THz spectroscopy.Firstly,the transmission mode of THz spectrometer was used to collect the time-domain data of samples.Secondly,the frequency domain data,refractive index data and absorption coefficient data were calculated.Then,the frequency domain spectrum,refractive index spectrum and absorption coefficient spectrum of vegetable leaf samples and soil samples were analyzed.The experimental results showed that when the Hg content of vegetable leaf and soil samples varied in the range of 0.00898-0.139 mg/kg and0.062-11.9 mg/kg,respectively,the refractive index showed a linear variation at 0.1975 THz,and the determination coefficients were 0.9736 and 0.9525,respectively.When the Cd content of leaf samples and soil samples varied in the range of 0.182～6.10 mg/kg and0.040～3.89 mg/kg,respectively,the refractive index showed a linear change at 0.3342 THz,and the determination coefficients were 0.9656 and 0.9726,respectively.When the Cu content of vegetable leaf and soil samples varied in the range of 3.30～30.7 mg/kg and37～320 mg/kg,respectively,the refractive index showed a linear change at 0.2672 THz,and the determination coefficients were 0.9845 and 0.8993,respectively.(2)Based on THz spectroscopy,the types of heavy metal pollution in the leaves of cabbage,the types of heavy metal pollution in the soil and the levels of heavy metal pollution in the soil were qualitatively identified.The THz time domain and frequency domain data of the sample were de-noised and dimensioning was reduced respectively,and then identification modeling was carried out.Wavelet Transform(WT),Standard Normal Variate Transformation(SNV)and WT-SNV were used to remove noise.Principal Component Analysis(PCA)and Multiple Dimension Scaling(MDS)methods were used to reduce dimensions.The recognition models are Probabilistic Neural Network(PNN),Random Forest(RF)and Deep Neural Network(DNN).The experimental results show that THz time domain data can better reflect the characteristics of heavy metals in soil and vegetable leaves than frequency domain data,and it is more suitable for identification and modeling.In the qualitative identification of vegetable leaf pollution types,the DNN identification model based on the THz time domain spectral data preprocessed by WT-SNV-MDS was the best,and the accuracy was 99.10%.In the identification of soil heavy metal pollution types,the PNN recognition model was constructed based on the THz time domain spectral data pretreated by WT-SNV.The recognition ability of the model was the best,and the accuracy was 99.22%.In the identification experiment of soil pollution levels of Hg,Cd and Cu,the WT-SNV-PCA-DNN model,WT-SNV-PNN model and WT-PCA-RF model have better recognition ability,and their recognition accuracy rates are99.83%,99.74% and 99.31%,respectively.(3)The contents of Hg,Cd and Cu in leaves and soil of Chinese cabbage were quantitatively predicted based on THz spectroscopy.The THz time domain and refractive index data of the samples were de-noised and dimension-reduced respectively,and then the prediction modeling was carried out.The denoising methods are WT and Discrete Cosine Transform(DCT),and the dimensionality reduction methods are PCA and Linear Discriminant Analysis(LDA).The prediction models are Extreme Learning Machine(ELM)and Back Propagation Neural Networks(BPNN).The experimental results show that THz time domain data is more suitable for heavy metal content prediction modeling than refractive index data.In the experiment of Hg content prediction of vegetable leaves,the BPNN model established after THz time domain data pretreated by DCT-PCA has the strongest predictive ability,and the root mean square error and determination coefficient are0.0072 and 0.9895,respectively.In the experiment of Cd content prediction of vegetable leaves,the BPNN model established after THz time domain data preprocessing by WT-PCA had the highest prediction accuracy,and the root mean square error and determination coefficient were 0.1355 and 0.9974,respectively.In the experiment of Cu content prediction of vegetable leaves,the BPNN model based on THz time domain data pretreated by DCT-LDA had the highest prediction accuracy,and the root mean square error and determination coefficient were 1.3763 and 0.9933,respectively.In the prediction experiment of soil Hg content,the BPNN model based on THz time domain data pretreated by DCT-PCA has the strongest prediction ability,and the root mean square error and determination coefficient are 0.2229 and 0.9940,respectively.In the prediction experiment of soil Cd content,the BPNN model established after THz time domain data pretreated by WT-PCA has the strongest prediction ability,and the root mean square error and determination coefficient are 0.0421 and 0.9990,respectively.In the prediction experiment of soil Cu content,the BPNN model established after THz time domain data pretreated by WT-PCA has the strongest prediction ability,and the root mean square error and determination coefficient are 4.9297 and 0.9981,respectively.