Detection of Total Phosphorus Concentration in Seawater Based on SERS Spectroscopy and Adaptive Locally Weighted Visible Image Combined with Parallel Self Weighted Graph Neural Network

Seawater eutrophication is a major global challenge for marine ecosystems, and accurate determination of total phosphorus (TP) concentration is crucial for monitoring and mitigating eutrophication. However, traditional detection methods often suffer from high costs, operational complexity, or limited detection sensitivity. To address these limitations, this study proposes a method for detecting TP in seawater based on surface-enhanced Raman spectroscopy (SERS) and graph neural networks (GNN). First, rhodamine 6 G (R6G)-modified silver nanoparticles (AgNPs) were utilized as the SERS active substrate to indirectly invert the total phosphorus concentration by combining the chemical products of phosphate ions and ammonium molybdate. Then, an Adaptive Local Weighted Visibility Graph (ALWVG) method was applied to transform SERS data, enhancing both local and global spectral features. Furthermore, a Parallel Self-Weighted Graph Neural Network (PSWGNN) was constructed, incorporating spectral chemical feature partitioning to achieve high-precision TP concentration prediction. Experimental results demonstrate that the proposed method achieves excellent detection performance in the 0–100 µg/L concentration range, with the R² of 0.996, the RMSE of 1.652 µg/L, the MAE of 1.138 µg/L, and the recoveries ranged from 97.20%-102.77%. Compared to traditional methods, this approach offers higher sensitivity, broader applicability, and better adaptation to complex aquatic environments. The findings suggest that the TP detection system based on SERS and ALWVG-PSWGNN provides an efficient, low-cost, accurate, and scalable solution for marine eutrophication monitoring.