Extracting Optical Fiber Background from Surface-Enhanced Raman Spectroscopy Spectra Based on Bi-Objective Optimization Modeling

Remote SERS detection at a 10-m scale using silica fiber SERS probes coupled with a convolutional neural network

A silica fiber surface-enhanced Raman scattering (SERS) probe provides a practical way for remote SERS detection of analytes, but it faces the major bottleneck that the relatively large Raman background of silica fiber itself greatly limits the remote detection sensitivity and distance. In this article, we developed a convolutional neural network (CNN)-based deep learning algorithm to effectively remove the Raman background of silica fiber itself and thus significantly improved the remote detection capability of the silica fiber SERS probes. The CNN model was constructed based on a U-Net architecture and instead of concatenating, the residual connection was adopted to fully leverage the features of both the shallow and deep layers. After training, this CNN model presented an excellent background removal capacity and thus improved the detection sensitivity by an order of magnitude compared with the conventional reference spectrum method (RSM). By combining the CNN algorithm and the highly sensitive fiber SERS probes fabricated by the laser-induced evaporation self-assembly method, a limit of detection (LOD) as low as 10^-8 M for Rh6G solution was achieved with a long detection distance of 10 m. To the best of our knowledge, this is the first report of remote SERS detection at a 10-m scale with fiber SERS probes. As the proposed remote detection system with silica fiber SERS probes was very simple and low cost, this work may find important applications in hazardous detection, contaminant monitoring, and other remote spectroscopic detection in biomedicine and environmental sciences.

Fitting an Optical Fiber Background with a Weighted Savitzky-Golay Smoothing Filter for Raman Spectroscopy

The Raman background arising from optical fiber materials poses a critical problem for fiber optic surface-enhanced Raman spectroscopy (SERS). A novel filter is developed to fit the optical fiber background from the measured SERS spectrum of the target sample. The general model of the filter is built by incorporating a weighted term of matching the similarity between the estimated background spectrum and the measured background spectrum into the classic Savitzky-Golay (SG) smoothing filter model. Through respectively selecting Euclidean cosine coefficient (ECos) and Pearson correlation coefficient (PCor) as the similarity index, two different models of the weighted SG smoothing filter are derived and named as SG-ECos and SG-PCor accordingly. Furthermore, the algorithm is presented, implemented, successfully applied to experimentally measured SERS spectra of rhodamine 6G and crystal violet, and validated with mathematically simulated Raman spectra. Experimental and simulation results show that the SG-ECos filter is effective, fast, flexible, and of certain anti-noise capability in background fitting. It is suggested that the proposed filter may be also applicable for other Raman spectra measurements to remove spectral contaminants originated from sampling substrates such as glass slides.

Microstructured Waveguides with Polyelectrolyte-Stabilized Gold Nanostars for SERS Sensing of Dissolved Analytes

A sensor based on microstructured waveguides (MWGs) with a hollow core inner surface covered with polyelectrolyte-layer-stabilized gold nanostars was developed for the SERS sensing of dissolved analytes. A polyelectrolyte-layer coating over the inner surface of glass cladding served as a spacer, reducing nonlinear optical effects in the glass near plasmonic hotspots of nanoparticles, as a stabilizing agent for thermodynamically unstable gold nanostars and as an optical coating for the fine-tuning of MWG bandgaps. This approach can be used to construct different kinds of SERS sensors for dissolved analytes, providing conservation, the prevention of coagulation, and the drying of a liquid sample for the time required to record the signal.