Rapid quantification of trace explosives with SERS using machine leaning and novel hydrophobic plasmonic paper substrate Reshma Beeram, Dipanjan Banerjee, Lingamurthy Narla, Soma Venugopal Rao * Advanaced Center for Research in High Energy Materials (ACRHEM), University of Hyderabad, Telangana * soma_venu@uohyd.ac.in OR soma_venu@yahoo.ac.in Surface Enhanced Raman Spectroscopy (SERS) is both a quantitative and qualitative technique to uniquely identify the analyte under study. Quantitative detection with SERS is a challenge owing to signal variation for inherent reasons like inhomogeneous distribution of hotspots, molecule orientation, non-uniform adsorption. Hence, simple linear models do not work because the concentration and intensity relation are often nonlinear and indirect [1] . Machine learning models are convenient for their ability to capture complex patterns in a given data set. However, the work done so far in quantification in SERS has relied on low RSD, commercial or rigid substrates. Here, we present low-cost and flexible Hydrophobic Filter Paper (HFP) substrate coupled with machine learning for quantification in SERS using a portable Raman spectrometer. Hydrophobic substrates were proven to have an advantage in SERS by concentrating the analyte and nanoparticles to a small area, thus increasing the density of hotspots [2] . The wettability of filter paper is modified by a simple method of spin coating it with Silicon oil for the first time. Gold nanoparticles were prepared by femtosecond laser ablation of pure gold in water with a pulse duration of ~50 fs (800 nm). These nanoparticles were characterized by UV-Visible spectroscopy, FESEM and TEM. Further, this combination of HFP and Au nanoparticles was used as SERS substrate for detection of a dye, Crystal Violet (CV) and an explosive molecule, Picric Acid (PA) with different concentrations. Using a portable Raman spectrometer, we have collected 100 spectra for each concentration of the analytes thus sampling the substrate well enough. Initially, the nonlinear features were extracted by Kernel Principal Component Analysis to reduce the dimensionality of the data and hence the time and complexity of future models. Support Vector Regression which is also a non-linear model, is used for the case of regression. The principal components with variance greater than 90% were fed to the support vector regression algorithm. Regression metrics were used to estimate the performance of the models. The coefficient of determination (R 2 ) for the case of CV was found to be 0.9629 and for PA 0.9472 with a computation time of less than 10 s. This methodology combined with flexible substrate and portable Raman Spectrometer can serve as a framework for onsite quantification using SERS. Figure 1 shows schematic of the summary of our work.
Figure 1: Schematic of Quantification in SERS using flexible plasmonic HFP combined with Machine Learning. References 1. Lussier, V. Thibault, B. Charron, G. Q. Wallace and J. F. Masson, Deep learning and artificial intelligence methods for Raman and surface-enhanced Raman scattering, TrAC - Trends Anal. Chem., 2020, 124, 115796. 2. De Angelis, F., et al., Breaking the diffusion limit with super-hydrophobic delivery of molecules to plasmonic nanofocusing SERS structures, Nat. Photonics, 2011, 5, 682–687.
V06
© The Author(s), 2022
Made with FlippingBook Learn more on our blog