Research Progress, Key Challenges, and Future Prospects of Raman Spectroscopy Combined With Machine Learning in Food Analysis

Surface-enhanced Raman spectroscopy (SERS) has emerged as an important technical method for food quality and safety detection owing to its advantages, such as molecular fingerprint specificity and rapidity without damage. However, complex food matrices, susceptibility of SERS signals to fluorescence, noise, scattering interference, redundant high-dimensional spectral data, and limited model generalization ability hinder the direct application of this technique to on-site industrial detection. In recent years, machine learning (ML) and deep learning have provided revolutionary solutions for signal correction, feature extraction, qualitative and quantitative analyses, and model transfer in SERS, achieving a full-chain upgrade from data collection to intelligent analysis. This paper begins with the four-layer relationships of instrument principle, signal mechanism, algorithm logic, and food system characteristics, systematically reviews the technical paradigm of the integration of SERS and ML, analyzes the real pain points and scientific bottlenecks currently faced, and proposes future development paths from the perspectives of basic mechanisms, model interpretability, industrialization implementation, and multimodal intelligent systems, aiming to provide systematic academic references for conducting high-depth and practical SERS-ML research in the field of food science.