Transformative role of deep learning in Raman spectroscopy-based detection of microplastics and nanoplastics

The pervasive presence of microplastics and nanoplastics (MNPs) in aquatic, terrestrial, and atmospheric systems has emerged as an important environmental challenge. Accurate detection and classification of these particles are essential for understanding their sources, fate, and potential risks to ecosystems and human health. Raman spectroscopy has become a leading analytical tool for MNPs characterization owing to its molecular specificity, non-destructive nature, and high spatial resolution. However, conventional Raman analysis often faces difficulties such as spectral noise, fluorescence interference, and the time-consuming nature of manual spectral interpretation. Recent advances in deep learning (DL) have introduced powerful means of enhancing Raman-based analysis by enabling automated, high-throughput, and data-driven spectral analysis. Nevertheless, many DL-based approaches have been validated primarily using controlled laboratory datasets, and their robustness in complex environmental matrices remains limited. This review provides a comprehensive overview of the integration of DL with Raman spectroscopy for the detection and quantification of MNPs. It summarizes current methodological developments, as well as perspectives on hybrid approaches that combine Raman data with complementary analytical techniques. Key achievements are discussed alongside persisting limitations, including domain shifts between reference and environmental spectra, uncertainty in ground-truth labeling, instrumental and inter-laboratory variability, and challenges related to data standardization and model interpretability. Rather than presenting DL as a universal solution, this review adopts a balanced and application-aware perspective, highlighting emerging directions such as transfer learning, self-supervised and federated approaches, and portable artificial intelligence-enhanced Raman platforms for real-time environmental monitoring, while emphasizing the current constraints of real-world deployment. • Overview of deep-learning integration with Raman spectroscopy for MNPs detection • Summary of neural networks usage in MNPs classification and quantification tasks • Discusses limitations in data standardization, model explainability, and scalability • Highlights emerging AI trends such as transfer and self-supervised learning • Outlines future steps toward portable, real-time AI-enhanced Raman platforms