Edge-Intelligent Surface-Enhanced Raman Spectroscopy Analytics for Organic Contaminants Detection in Coastal Marine Environments

Marine pollution poses a critical global environmental challenge, exacerbated by accelerating industrialization and expanding human activities. The discharge of pollutants threatens marine ecosystem stability, biodiversity, human health, and coastal socio-economic sustainability. Traditional laboratory-based pollutant detection methods are often cumbersome, costly, and lack real-time field capabilities. To address these issues, we propose an edge-intelligent framework called MariEdge-SERS for on-site marine pollutant detection integrating Surface-Enhanced Raman Spectroscopy (SERS) and Graph Neural Networks (GNNs). It employs portable Raman spectrometers for field-based spectral data acquisition. GNN models implemented on edge computing platforms perform real-time, high-fidelity spectral analytics at potential pollution sources, with analytical results transmitted to Cloud servers for archival storage and advanced interpretation. The main innovations of our proposal include: (1) A closed-loop “collection-analytics-feedback” pipeline enabling real-time field detection; (2) a graph-based spectral representation that significantly enhances identification and prediction accuracy through GNNs; and (3) edge deployment ensuring efficient operation in resource-constrained environments. Experimental results indicate that MariEdge-SERS exhibits superior accuracy and robustness in both qualitative identification and quantitative prediction, achieving 100% and 98.3% accuracy, respectively. Furthermore, the Limit of Detection (LOD) of MariEdge-SERS is 0.03 µM, demonstrating the high sensitivity of our method. This showcases its potential application in marine pollution monitoring and provides new insights for intelligent field-deployable environmental monitoring.