We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Study on Rapid Quantitative Detection of Soil MPs Based on Terahertz Time-Domain Spectroscopy
Summary
Researchers developed a rapid method for detecting and quantifying microplastics in soil using terahertz time-domain spectroscopy combined with machine learning algorithms. The classification models achieved high accuracy in identifying different types of microplastics including polyethylene, polystyrene, and polypropylene. The study suggests that terahertz spectroscopy could provide a faster and more efficient alternative to current methods for monitoring microplastic contamination in agricultural soils.
The presence of microplastics (MPs) in agricultural soils substantially affects the growth, reproduction, feeding, survival, and immunity levels of soil biota. Therefore, it is crucial to investigate fast, effective, and accurate techniques for the detection of soil MPs. This work explores the integration of terahertz time-domain spectroscopy (THz-TDS) techniques with machine learning algorithms to develop a method for the classification and detection of MPs. First, THz spectral image data were preprocessed using moving average (MA). Subsequently, three classification models were developed, including random forest (RF), linear discriminant analysis, and support vector machine (SVM). Notably, the SVM model had an F1 score of 0.9817, demonstrating its ability to rapidly classify MPs in soil samples. Three regression models, namely, principal component regression (PCR), RF, and least squares support vector machine (LSSVM), were developed for the detection of three MPs polymers in agricultural soils. Six feature extraction methods were used to extract the relevant parts of the data containing key information. The results of the study showed that the regression accuracies of PCR, RF, and LSSVM were greater than 83%. Among them, the RF had the highest overall regression accuracy. Notably, PE-UVE-RF had the best performance with Rc2, Rp2, root mean square error of calibration, and root mean square error of prediction values of 0.9974, 0.9916, 0.1595, and 0.2680, respectively. Furthermore, this model gets a better performance by hypothesis testing and predicting real samples.