Interfacial Regulation-Driven Dual-Enrichment SERS Coupled With Deep Learning Enable Ultrasensitive and In Situ Identification of Microplastics in Natural Waters.

Accurate detection of trace-level microplastics in natural waters remains challenging due to inefficient particle enrichment, unstable particle-substrate interactions, inhomogeneous hotspot distributions, and spectral similarity among polymers. Here, we develop a membrane-confined SERS platform integrated with machine-learning-assisted spectral analysis for sensitive and reproducible microplastic detection. In this platform, PSS-Na-modified Au@Ag nanocubes are employed as plasmonic building blocks to construct a charge-regulated, vertically stacked hotspot architecture on a cellulose acetate membrane. By improving microplastic enrichment, hotspot accessibility, and interlayer electromagnetic coupling, the membrane-guided configuration enables reproducible detection of polystyrene down to 50 ng mL- 1. It also delivers highly reproducible signals in microplastic spike samples prepared in six environmental water matrices without chemical pretreatment. To resolve spectral similarity among polymers, a multi-polymer SERS library covering PS, PMMA, PVC, and PC was integrated with an attention-based 1D-CNN, enabling reliable polymer identification and semi-quantitative analysis of binary and ternary mixtures (R2 > 0.83). This membrane-confined plasmonic platform, together with data-driven spectral analysis, provides a robust route for intelligent microplastic monitoring in complex aquatic environments.