# Synergistically Enhanced Ta2O5/AgNPs SERS Substrate Coupled with Deep Learning for Ultra-Sensitive Microplastic Detection

## TL;DR

This paper introduces a highly sensitive method for detecting microplastics using a Ta2O5/AgNPs SERS substrate and a deep learning model, achieving ultra-low detection limits and high accuracy.

## Contribution

A novel Ta2O5/AgNPs SERS substrate and a CNN-Transformer model for ultra-sensitive and accurate microplastic detection in complex environments.

## Key findings

- Ta2O5/AgNPs substrates achieved a detection limit of 10−13 M for R6G.
- The CNN-Transformer model achieved 98.7% accuracy in high-noise spectral analysis.
- The pseudo-Neuston network effectively captures microplastics like PS, PET, and PMMA.

## Abstract

What are the main findings?
Fabricated spherical Ta2O5/AgNPs substrates with pseudo-Neuston networks.Achieved an ultra-low detection limit of 10−13 M for R6G via EM/CM contribution.Developed a CNN-Transformer model achieving 98.7% accuracy in high-noise spectra.

Fabricated spherical Ta2O5/AgNPs substrates with pseudo-Neuston networks.

Achieved an ultra-low detection limit of 10−13 M for R6G via EM/CM contribution.

Developed a CNN-Transformer model achieving 98.7% accuracy in high-noise spectra.

What are the implications of the main findings?
Provides a scalable strategy for enhancing semiconductor SERS activity.Overcomes spectral interference in complex environmental microplastic detection.Demonstrates deep learning’s potential in robust automated spectral analysis.

Provides a scalable strategy for enhancing semiconductor SERS activity.

Overcomes spectral interference in complex environmental microplastic detection.

Demonstrates deep learning’s potential in robust automated spectral analysis.

Herein, a high-performance Ta2O5/AgNPs composite Surface-Enhanced Raman Scattering (SERS) substrate is engineered for highly sensitive detection of microplastics. Through morphology modulation and band-gap engineering, the semiconductor Ta2O5 is structured into spheres and composited with silver nanoparticles (AgNPs), facilitating efficient charge transfer and localized surface plasmon resonance (LSPR). This architecture integrates electromagnetic (EM) and chemical (CM) enhancement mechanisms, achieving an ultra-low detection limit of 10−13 M for rhodamine 6G (R6G) with excellent linearity. Furthermore, the three-dimensional “pseudo-Neuston” network structure exhibits superior capture capability for microplastics (PS, PET, PMMA). To address spectral interference in simulated complex environments, a multi-scale deep-learning model combining wavelet transform, Convolutional Neural Networks (CNN), and Transformers is proposed. This model achieves a classification accuracy of 98.7% under high-noise conditions, significantly outperforming traditional machine learning methods. This work presents a robust strategy for environmental monitoring, offering a novel solution for precise risk assessment of microplastic pollution.