Detection Of PAHs By Surface-Plasmon Enhanced Raman Spectroscopy Combined With Deep Learning

Polycyclic aromatic hydrocarbons(PAHs)are persistent organic pollutants composed of multiple aromatic rings in different configurations.They are widely distributed in atmosphere,soil,water and plant,and accumulate in foods such as vegetables,fruits and grains through above media,and finally enter the human body through food intake,respiratory intake and skin exposure,causing carcinogenic,mutagenic,immunosuppressive and teratogenic damage.Therefore,it is very meaningful to detect PAHs residues in the human body and foods.In order to develop a simple,accurate,intelligent and real-time quantitative and qualitative detection method for trace PAHs residues,surface-plasmon enhanced Raman spectroscopy(SERS)was used in combination with surface-functionalized nanoparticles,liquidliquid interface self-assembly(LLISA),superhydrophobic platform design to realize the high sensitive and stable detection of PAHs residues in human urine and foods.Meanwhile,the deep learning(DL)method was combined with SERS spectroscopy to construct a classification and regression model to realize the rapid and intelligent detection and analysis of PAHs residues.The main research contents are as follows:(1)Preparation and response to PAHs of plasmon nanoparticles.Nanoparticles with different morphologies were prepared and their electromagnetic field distribution was simulated.The response of nanoparticles functionalized with different surface modifiers to PAHs was explored.The properties and trapping mechanism with PAHs of the modified molecules with the best response were analyzed.The results showed that spherical nanoparticles with tips and sharp gaps had strong electromagnetic field,which was helpful to enhance SERS signal.β-cyclodextrin surface functionalized gold nanoparticles(β-CD@AuNPs)showed the best response to PAHs.PAHs molecules were trapped by β-CD with inherent hydrophobic cavities to form complexes under host-guest interaction,which enhanced the adsorption to PAHs and improved the sensitivity.Moreover,the solution reached a stable system after mixing PAHs with βCD@AuNPs,which could realize SERS detection with high reproducibility.(2)Detection of PAHs residue on urine and fruit and vegetable surface by SERS based on β-CD@AuNPs.The effect of urine pretreatment method on SERS detection of OH-PAHs was analyzed.The influence of nano-structure assembly on SERS detection performance and the response of different SERS detection methods to OHPAHs were explored.The results showed that when β-CD@AuNPs was used as SERS substrate,the lowest concentrations of 1-ohpyrene(1-OHPyr),2-hydroxy-naphthalene(2-OHNap)and 3-hydroxy-benzopyrene(3-OHBap)in urine were 0.5 μg/mL,0.5μg/mL and 0.1 μg/mL,with relative standard deviation(RSD)of 7.8%,8.2%and 7.0%,respectively,which indicating good sensitivity and stability.Subsequently,in order to further improve the sensitivity,the monolayer β-CD@AuNPs assembly films were prepared using LLISA method,which produced a large number of hot spots.The contact probability between molecules and hot spots was improved by injecting OHPAHs under the wet assembled film to construct the nanocapillary pump model.The detection limit of 1-OHPyr in urine was 0.05 μg/mL,and the RSD was 5.5%,realizing highly sensitive and stable SERS detection of OH-PAHs in urine.Finally,the hydrophobic smooth flexible platform was constructed by coating the surface of the polytetrafluoroethylene film(PFTE)with perfluorinated liquid.β-CD@AuNPs on the PFTE were concentrated under hydrophobic action to form a compact and uniform nanoarray with high sensitivity and stability.The results show that βCD@AuNPs/PFTE platform has excellent response and applicability.The low detection concentrations of PAHs residues such as Bap,Pyr,Nap on the surface of fruits and vegetables were 0.25 μg/cm2,0.5 μg/cm2 and 0.25 μg/cm2,which realized in situ,rapid and non-destructive detection of PAHs residues on irregular food surface.(3)Qualitative/quantitative intelligent analysis detection of PAHs was realized through SERS spectrum combined with DL to construct a classification/regression model.First,the PAHs classification recognition model was constructed by using the Inception network,and the improvement of Inception network performance by adding residual and attention models was explored.The results showed that the model built with Inception-residual-attention had the best prediction performance,with the accuracy rates of 98.98%,94.00%and 98.13%for the training set(ACCT),calibration set(ACCV)and prediction set(ACCP),respectively.Subsequently,a regression model based on Inception network was established to realize the quantitative analysis of PAHs,and the optimal model was sought by comparing with several classical ML and DL methods.The predictive performance of the model combined with feature extraction was also explored.The results showed that the convolutional neural network and genetic algorithm obtained the best prediction results,the coefficient of determination and root mean square error of the prediction set were 0.9639 and 0.6327,respectively,which could realize the intelligent analysis and accurate prediction of PAHs.Finally,the feasibility of rapid intelligent identification of mixed PAHs residues was studied.In order to ensure the accuracy of the model,improve the computing speed and reduce the consumption of computer resources,the recognition capability of lightweight deep network was explored.The results showed that the ShuffleNet network has the best recognition effect with ACCT,ACCV and ACCP of 100%,96.61%and 97.63%,respectivily,realizing the classification of multiple mixed PAHs.In conclusion,this paper realized sensitive,rapid and accurate quantitative and qualitative detection of PAHs residue in urine and the surface of fruit and vegetable through SERS combined with DL.