Fluorescence spectral identification of soil microplastics based on FA-IGWO-ELM

Microplastics are widespread contaminants in soil environments, creating a need for rapid and accurate methods for polymer identification. Fluorescence spectroscopy offers a sensitive, low-cost and nondestructive analytical route, but spectral overlap among soil–microplastic samples limits the reliability of visual inspection and single-peak analysis. Here, we propose a factor-analysis-assisted Extreme Learning Machine optimized by an Improved Grey Wolf Optimizer (FA-IGWO-ELM) for the fluorescence spectral identification of soil microplastics. Moving-average filtering was used for noise reduction, factor analysis was applied for dimensionality reduction, and IGWO was used to optimize the input weights and hidden-layer biases of ELM. The IGWO algorithm incorporated chaotic initialization, a population mean-guided search strategy and a greedy selection mechanism to improve optimization stability. The method was evaluated on a main classification dataset containing seven common microplastic polymers, namely PE, PP, PVC, PS, PA, PET and PC, under a stratified five-fold cross-validation framework. Compared with FA-SVM, FA-XGBoost, FA-ELM, FA-GWO-ELM and FA-ACO-ELM, the proposed FA-IGWO-ELM model achieved the best overall performance, with a mean accuracy of 97.14%, a macro-recall of 0.9714 and a macro-F1-score of 0.9715. The AUC values for the seven polymer categories ranged from 0.988 to 1.000, indicating strong class-discrimination ability. A targeted robustness validation using two soil matrices and three particle-size conditions further yielded an overall accuracy of 97.22%, supporting the stability of the method under controlled variations in sample conditions. These results suggest that fluorescence spectroscopy combined with FA-IGWO-ELM is a feasible approach for rapid and accurate identification of soil microplastics.