Deep Learning for Microplastic Detection and Characterization Across Imaging, Spectroscopic, and Multimodal Platforms: Advances, Validation Gaps, and Deployment Challenges

Microplastics are now recognized as a persistent and heterogeneous class of environmental contaminants, whose reliable monitoring requires analytical strategies capable of handling large numbers of particles across complex matrices. Although Fourier-transform infrared and Raman spectroscopy remain reference techniques for polymer identification, their routine application is often constrained by time-consuming sample processing, manual interpretation, and limited scalability. Deep learning is increasingly being explored as a way to transform microplastic analysis from operator-dependent workflows into automated, reproducible, and high-throughput pipelines. This review critically examines recent advances in deep learning for microplastic detection, classification, and characterization across imaging, spectroscopic, and multimodal platforms. A structured literature search was conducted using the Web of Science Core Collection, focusing primarily on peer-reviewed studies published between 2016 and 2026 and selecting representative contributions according to their methodological relevance, validation design, environmental applicability, and deployment potential. Imaging-based approaches are discussed in relation to particle localization, segmentation, and counting, while spectroscopic models are evaluated for polymer-specific classification and aging assessment. Particular emphasis is placed on multimodal architectures that combine morphological and chemical information, with the potential to improve analytical robustness. Rather than focusing only on reported accuracy, this review considers the practical barriers that currently limit real-world implementation, including dataset scarcity, domain shift between controlled and environmental samples, spectral variability, limited model interpretability, and computational requirements for portable or in situ deployment. By comparing methodological trends across different analytical platforms, this work highlights the gap between proof-of-concept demonstrations and operational monitoring systems. Finally, it outlines key priorities for future research, including standardized datasets, transparent validation protocols, explainable models, domain-adaptation strategies, and lightweight architectures suitable for field applications.