An all-in-one analytical platform for rapid determination of toxic dyes in complex water by SERS and machine learning.

Rapid and reliable determination of toxic dyes in complex aqueous environments remains challenging because matrix interference and complicated sample pretreatment often compromise analytical accuracy and practical applicability. Therefore, an integrated filtration-adsorption-SERS sensing platform coupled with one-dimensional convolutional neural network (1D-CNN) modeling was fabricated for rapid and quantitative detection of malachite green (MG) and methylene blue (MB) in complex water samples. Polypropylene filter cloth (PPFC) served as a porous fibrous support for sample filtration, while Ag nanoparticles (Ag NPs) were uniformly immobilized on the fiber surfaces via a seed-mediated in situ growth strategy to provide SERS-active sites. Activated carbon was incorporated into the PPFC/Ag filter configuration to provide adsorption of the target dyes during sample treatment. The resulting configuration integrates filtration, adsorption, and SERS detection within a common analytical workflow, while 1D-CNN modeling was employed for quantitative analysis of the acquired SERS spectra. Benefiting from the integrated filtration-adsorption configuration and porous fibrous architecture, the platform exhibited high signal uniformity and favorable analytical performance, with detection limits of 10-9 M for both MG and MB. Adsorption kinetic analysis demonstrated that dye removal followed the pseudo-second-order model, with equilibrium removal efficiencies of 99.7% for MB and 98.6% for MG. To improve quantitative reliability in complex matrices, a 1D-CNN model was employed for SERS spectral analysis, yielding high accuracy with R2 values of 0.992 and 0.982 for MB and MG, respectively, with low prediction errors. Furthermore, spiked Moyan Lake water samples showed recoveries of 98.14% and 96.37% for MB and MG, respectively, with RSD values of 4.53% and 4.91%, demonstrating satisfactory accuracy and precision in a real water matrix. By integrating these complementary functions within the same filtration configuration, the proposed platform establishes a continuous analytical workflow from sample filtration and adsorption to SERS detection and quantitative analysis. This integrated design simplifies the conventional multi-step procedure and provides a sensitive, reliable, and practical approach for rapid monitoring of trace organic dyes in complex water matrices.