Data Fusion Enhanced High-Dimensional SERS Fingerprints Construction via Dual-Wavelength and Multisubstrate Strategy for Precise Wastewater Identification.

It is well-known that traditional label-free surface-enhanced Raman spectroscopy (SERS) can capture fingerprint information on analyte, providing a foundation for target identification and differentiation. However, the conventional one-dimensional spectral data obtained through traditional SERS methods is insufficient for characterizing samples with complex chemical compositions, such as wastewater, or for tackling more intricate challenges, including tracing pollution sources, where a more comprehensive analytical profile is necessary. Herein, we introduce "SERSynergy", a data-fusion-driven machine learning approach that integrates dual-wavelength and multisubstrate data to generate a holistic SERS fingerprint, which allows for precise and robust wastewater identification. This method leverages complementary spectral features of wastewater samples by collecting a total of 12,000 spectra using four types of noble metal nanoparticles under two excitation wavelengths. A hybrid feature-decision fusion strategy cross-combined spectral features from various conditions to form high-dimensional fingerprints, which were then evaluated using optimized machine learning models and consolidated via probability-level fusion. The "SERSynergy" method demonstrated an identification accuracy of up to 99.67% for wastewater samples. Furthermore, when validated with blind sample testing, the method maintained an accuracy of 96.67%. Overall, the developed approach shows great promise for efficiently and accurately identifying wastewater samples, and it has potential applications in the precise acquisition of spectral features and identity discrimination in complex matrix samples.