Detection of prohibited drugs in aquatic products using surface-enhanced Raman spectroscopy combined with a CNN-DE-LSTM deep learning model

Xijun Wu, Shixin Wang, Xuan(轩) Zhao, Jiangtao Wang, Xuan(煊) Zhao, Yungang Zhang,
Detection of prohibited drugs in aquatic products using surface-enhanced Raman spectroscopy combined with a CNN-DE-LSTM deep learning model,
Microchemical Journal,
Volume 219,
2025,
116035,
ISSN 0026-265X,
https://doi.org/10.1016/j.microc.2025.116035.
(https://www.sciencedirect.com/science/article/pii/S0026265X25033831)
Abstract: In this study, we constructed a surface-enhanced Raman substrate based on Ag@COF nanostructures of Paranucleo-Choco Tegmen, combined with a convolutional neural network-dynamic equilibrium-long short-term memory (CNN-DE-LSTM) deep learning model optimized through a dynamic equilibrium (DE) mechanism, and established a new method for trace detection of prohibited drugs (MG/CAP/NOR) in aquatic products. Mechanistic studies showed that the synergistic effect of Ag nanoparticles and covalent organic frameworks (COFs) substrate could generate a 1.04 × 108 EMF enhancement factor and significantly expand the hot spot region. In qualitative detection, the Gradient Boosting Decision Tree (GBDT) model achieved 98 % accuracy; in quantitative analysis, the coefficients of determination (R2) of the CNN-DE-LSTM model for all three drugs exceeded 0.998, and the detection sensitivity in real samples was up to the pM level (MG: 7.94 × 10−12), with recovery rates ranging from 89.25 % to 95.00 %. Compared with traditional CNN and CNN-LSTM, the convergence speed of this model is improved by 40 %, and the feature extraction ability is significantly enhanced. The method breaks through the bottleneck of trace detection under the interference of complex matrix and provides an innovative solution for intelligent monitoring of food safety.
Keywords: SERS; Aquatic products; Prohibited drugs; CNN-DE-LSTM