We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Detection and cross-organ characterization of physiological response to microplastic stress in Panax ginseng based on hyperspectral imaging assisted with machine learning
Summary
Scientists found that microplastic pollution stresses out ginseng plants, causing measurable changes in their chlorophyll, sugars, and antioxidant levels—key markers of the plant's health and medicinal quality. Using special camera imaging and AI, they could detect this stress just by scanning the leaves, without digging up the roots, showing a fast way to check if popular herbal supplements are being grown in contaminated environments. This matters because it suggests microplastics in soil could be quietly affecting the quality of medicinal plants many people rely on for health remedies.
Microplastic pollution can affect growth and quality of medicinal plants, yet rapid detection of microplastic stress responses remains underexplored. We treated ginseng with polyethylene microplastics, acquired leaf hyperspectral images (HSI) on day 23, and constructed machine learning models for identifying stress levels and predicting physiological indicators. Furthermore, the applicability of successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS) for characteristic wavelength selection was compared. Results showed that polyethylene stress significantly affected the physiological state. The classification models effectively identified microplastic stresses of different concentrations, with the support vector machine (SVM) model performing the best (accuracy of 85.2%). For quantitative prediction, the partial least squares regression (PLSR) model exhibited optimal performance for indicators including chlorophyll (Chl) (RPD = 3.98), soluble sugar (RPD = 2.56) and peroxidase (POD) (RPD = 2.89), and the convolutional neural network performed better in superoxide dismutase (SOD) prediction (aerial RPD = 3.27, underground RPD = 2.65). Leaf spectral data enabled prediction of aerial and underground physiological indicators (RPD = 2.10 to 2.73), indicating that aerial spectral information reflected underground physiological state. Characteristic wavelength selection results showed that SPA had advantages for SOD prediction, while CARS performed better for the remaining seven indicators (RPD >2.0). In conclusion, HSI combined with machine learning models enabled rapid nondestructive identification of microplastic stress responses and prediction of key physiological indicators in ginseng, suggesting quantifiable relationships between aerial spectral data and underground physiological states. This study provides a technical prototype for the growth detection of medicinal plants.