We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Efficient screening of microplastics in soils using hyperspectral imaging in the short-wave infrared range coupled with machine learning – A laboratory-based experiment
Summary
Researchers tested short-wave infrared hyperspectral imaging combined with machine learning to detect three types of microplastics in soil, finding it could identify elevated contamination but was not sensitive enough for typical environmental background levels. The technique shows most promise for screening heavily polluted sites like landfills and industrial areas.
Microplastics (MP) in soil have emerged as an environmental pollutant of increasing interest in recent years, emphasizing the need for efficient screening methods. Hyperspectral imaging in the visible and near-infrared (VNIR) and short-wave infrared (SWIR) coupled with machine learning (ML) have shown potential for rapid, cost-effective MP detection in soils. However, key methodological challenges, including optimal ML algorithms for MP classification, detection limits dependent on MP type, and scaling relationships between area-based hyperspectral imaging and soil MP concentrations, should further be explored.In this study, we explored the potential of SWIR hyperspectral imaging to detect and quantify three MP types (polyamide − PA, polyethylene − PE, polypropylene − PP) at the (sub-)pixel level in soil-MP mixtures with concentrations ranging from 0.01 wt-% to 5.00 wt-% using Partial Least Squares − Discriminant Analysis (PLS-DA), Random Forests (RF), 1D-Convolutional Neural Networks (1D-CNN) and a three-model ensemble.All machine learning algorithms achieved comparable classification accuracies in a calibration–validation approach on a large spectral library developed from pure material spectra. When applied to the independent SWIR image data, RF performance decreased markedly, whereas the ensemble proved beneficial to suppress individual model-specific random misclassifications.We found a close non-linear relationship between the SWIR image area-based MP quantification and the actual concentration (wt-%) of MP in the soil samples that depended on the MP type. Interpolated MP detection limits were also MP-type specific and corresponded to 0.05 wt-%, 0.46 wt-% and 1.15 wt-% for PE, PP and PA, respectively, with the larger PE particles having a lower detection limit than the more finely dispersed PA and PP particles.Our results show that hyperspectral SWIR imaging has the potential to enable screening applications where elevated MP levels can occur, such as landfill or industrial sites, but is likely not sensitive enough to detect current background concentrations.