Microplastics (MPs), as ubiquitous environmental pollutants, pose significant ecotoxicological risks due to their accumulation in marine food chains and role as contaminant carriers. However, their rapid identification and aging assessment within complex marine matrices remain challenging. We established a Raman spectral dataset encompassing common polymer types across multiple artificial and natural aging stages. Using this dataset, an attentional neural network (aNN) trained under controlled laboratory conditions achieved high-precision classification of polymer types and aging states of natural aging MPs in seawater, reaching over 97% and 93% accuracy for polymer and aging state identification. To reduce interference from complex environmental matrices during MP identification in real seawater, an open-set deep learning (OSDL) framework was employed. This approach achieved an average identification accuracy of 94% for naturally aged MPs and 97% for nontarget particles and impurities in seawater, outperforming conventional closed-set algorithms, while field-based Validation using real microplastic fragments from coastal seawater yielded 90% accuracy. Collectively, these results demonstrate that Integrating Raman spectroscopy with the OSDL algorithm establishes a foundational approach for marine MP identification and aging characterization under semicontrolled conditions. Meanwhile, further development is required for diverse polymer formulations and geographic settings to enhance the universality of the approach.