Deep Learning–Based Classification of SERS Spectra for Simultaneous Identification of Clenbuterol, Ractopamine and Salbutamol

Abstract

The illicit residue of lean meat agents poses a severe challenge to global food safety, seriously threatening public health. To address the complex and time-consuming nature of traditional detection methods, this study proposes a novel classification method based on a deep learning neural network for Surface-Enhanced Raman Scattering (SERS) spectra, aiming to achieve rapid and accurate identification of three typical lean meat agents: clenbuterol, salbutamol, and ractopamine. Targeting the common issues of limited sample sizes and model overfitting in practical detection, we designed an optimized one-dimensional convolutional neural network (1D-CNN) architecture. Through the synergistic action of multi-layer convolutions and residual connections, the model can adaptively extract high-dimensional abstract features from complex SERS spectra, achieving precise classification via a fully connected layer. Experimental results demonstrate the method's outstanding performance, achieving an accuracy of 90.95% and an F1-score of 91.13% on an independent test set. In a systematic comparison with 13 traditional machine learning methods, including Random Forest and Support Vector Machines, the proposed method showed significant advantages across all performance metrics, particularly in generalization ability and stability, successfully suppressing the overfitting phenomenon common in small-sample datasets. This research not only provides an efficient and reliable new tool for the rapid screening of lean meat agents but also offers an important practical example and methodological reference for the application of deep learning in the analysis of small-sample spectral data.