Hyperspectral imaging and self-organizing map approach for non-destructive monitoring of microplastic contamination in sandy substrates

Abstract Microplastics (MPs) are a growing environmental concern, requiring effective methods for identification and quantification. This study develops and evaluates an application-oriented workflow combining near-infrared hyperspectral imaging (NIR-HSI) with a self-organizing map (SOM) and a percent-based expansion tolerance (PBET) for simultaneous microplastic mapping and semi-quantitative surface-coverage estimation. Spectral data were collected from MP fragments (PET, PE, PP, and PS; 1–5 mm) prepared from commercial household plastic source materials and experimentally distributed on sand surfaces at varying %coverage (0.78–12.5%). The NIR-HSI spectra were preprocessed to enhance spectral data quality. The modified SOMs successfully classified MPs, which were visually represented by distinct RGB color mappings. Qualitative results confirmed accurate visual identification of both individual and combined MPs, while quantitative results demonstrated strong predictive performance ( R ² up to 1.00 and low RMSE). Additionally, the robustness of the SOM model was evaluated under controlled conditions simulating real-world variability including particle size, pigment color, and overlapping MPs and further demonstrated using unknown MPs collected from natural beach samples. Notably, the approach also proved capable of detecting and classifying particles smaller than 1 mm, highlighting its high sensitivity for identifying small MPs that are often overlooked by conventional visual methods. The results demonstrate the potential of the NIR-HSI–SOM workflow for polymer-class mapping and semi-quantitative surface-coverage estimation of PET, PE, PP, and PS on the tested sand substrates, providing an analytical basis for polymer-specific assessment of MP contamination under the investigated conditions.