Identification of microplastics using a convolutional neural network based on micro-Raman spectroscopy

Microplastics (MPs) pose a threat to human and environmental health, and have emerged as a global environ-mental issue. Because MPs are small and complex, methods of quickly and reliably classifying and identifying them are either lacking or in the early stages of development. In this study, micro-Raman spectroscopy and a convolutional neural network (CNN) were combined to establish identification models for 10 MP references and three environmental samples. In addition, an interaction network was established based on pair-wise correlations of Raman bands to determine the influence of environmental stress on MPs. The CNN model achieved average classification accuracies of 96.43% and 95.6% for the 10 MP references and the three environmental samples, respectively. For MPs exposed to environmental stressors, an interaction network can provide highly sensitive, information-dense, and universally applicable signatures for characterizing the environmental processes affecting MP spectra. The results of this study can help establish efficient and automatic analysis for accurate identification of MPs as well as an intuitive exhibition of spectral changes on environmental exposure.