VNIR and SWIR Hyperspectral Imaging for Microplastic detection on Soil

Microplastics in soil significantly threatens ecology, impacting plant growth, soil, and humans health through the food chain. Conventional methods to detect microplastic in soil usually require complicated and time-consuming steps. This study used non-destructive hyperspectral imaging techniques in visible-near infrared (VNIR, 400-1000 nm) and short-wave-infrared (SWIR, 1000-2000) to identify microplastic in the soil surface. Seven cryo-milled microplastic polymer were used. Partial least squares discriminant analysis (PLS-DA), linear discriminant analysis (LDA), and support vector classification (SVC) with linear, polynomial, and radial basis function kernels were used to develop the calibration model. The result shows that in both VNIR and SWIR regions, models with linear kernel (PLS-DA, LDA, and SVC-linear) were superior to the non-linear model (SVC-poly and SVC-RBF). The masked image of SVC-linear model using VNIR SNV spectra was superior to the other VNIR model but could only differentiate microplastic from soil. The LDA model yield using the original SWIR spectra was performed perfectly, outperforming the other model with a clear classification of soil and each polymer in the masked validation image. This study provides initial insights into soil microplastic detection by hyperspectral imaging (HSI), presenting a practical, non-destructive method for the efficient identification of microplastic polymers without complicated sample preparation.