Accelerating microplastic analysis with machine learning: Utilizing multi-layer perceptron (MLP) for classification of hyperspectral data

Microplastic analysis using Fourier-transform infrared (FTIR) imaging plays a crucial role in understanding the distribution and composition of plastic pollution. However, traditional classification methods, such as library-based spectral correlation, are computationally demanding and require extensive manual validation, limiting their efficiency in large-scale studies. In this work, we present a semi-automatic workflow utilizing a Multi-Layer Perceptron (MLP) for the classification of hyperspectral FTIR data. Spectral data were extracted, preprocessed through absorbance conversion and baseline correction, and classified into 13 predefined categories, covering the 10 EU-priority polymers, the sample holder, natural polyamide and potassium bromide. A key feature of this approach is the integration of an expert-guided validation step, where classified spectra are clustered based on spatial proximity and compared to reference spectra. This allows for the identification and removal of potential misclassifications before final image analysis, ensuring a balance between automation and expert oversight. The proposed workflow significantly reduces analysis time, as traditional spectral correlation methods require several hours for classification, while the MLP-based system completes the process in 10–30 min per dataset. With an overall classification accuracy of 99.96 %, the results demonstrate the feasibility of using neural networks for microplastic identification. The approach enables efficient processing of large hyperspectral datasets measured in multiple subareas (“markers”) while maintaining high interpretability. Future work should focus on expanding the training dataset to enhance generalization and further refining the validation process. This study highlights the potential of machine learning for high-throughput microplastic analysis, providing a foundation for further advancements in AI-assisted spectral classification. • MLP-based method accelerates FTIR microplastic identification. • Achieved 99.96 % accuracy in microplastic classification. • Reduced analysis time from hours to 10–30 min per dataset. • Expert-guided validation enhances reliability of results. • Machine learning enables high-throughput microplastic analysis.