Deep learning assisted ATR-FTIR and Raman spectroscopy fusion technology for microplastic identification

Abstract

Microplastics, recognized as persistent environmental pollutants, have garnered significant global attention due to their widespread distribution and resistance to degradation. Accurate identification and classification of microplastics are crucial for monitoring pollution levels and assessing potential health risks. In this study, we employed Raman spectroscopy and Attenuated Total Reflection Fourier Transform Infrared Spectroscopy (ATR-FTIR) to analyze eight types of microplastics, establish a spectroscopic database. Additionally, we developed a one-dimensional convolutional neural network (1D-CNN) model that incorporates an embedded multi-head attention mechanism for the classification of eight kinds of microplastics. The recognition accuracy is 73% for ATR-FTIR and 75% for Raman. In order to overcome the shortcomings of single spectral data, a three-level data fusion classification and recognition algorithm were developed to leverage the complementary strengths of ATR-FTIR and Raman spectroscopy. The classification accuracies achieved by the low-level, mid-level, and high-level fusion models were 88%, 97%, and 99%, respectively. We further investigate the model’s applicability to real samples, by conduct spiked tests across three different media (milk, coke and tap water). The data obtained from these tests served as an external validation set to assess the model’s generalization ability. Notably, the recognition accuracy of the high-level fusion model exceeds 98% in all three spiked media. This study provides a more robust and effective method for the accurate classification and identification of microplastic.