Study on Rapid Quantitative Detection of Soil MPs Based on Terahertz Time-Domain Spectroscopy

Area of Science:

• Environmental Science
• Analytical Chemistry
• Spectroscopy

Background:

• Microplastics (MPs) negatively impact soil biota, necessitating efficient detection methods.
• Current techniques for soil MP analysis can be slow and labor-intensive.
• Developing rapid and accurate MP detection is crucial for agricultural soil health.

Purpose of the Study:

• To integrate terahertz time-domain spectroscopy (THz-TDS) with machine learning for classifying and detecting MPs in agricultural soils.
• To evaluate the performance of different machine learning models for MP classification and quantification.
• To establish a robust and efficient analytical method for soil microplastic analysis.

Main Methods:

• Preprocessing of THz spectral image data using moving average (MA).
• Development and comparison of classification models: Random Forest (RF), Linear Discriminant Analysis, Support Vector Machine (SVM).
• Development and evaluation of regression models (Principal Component Regression (PCR), RF, Least Squares Support Vector Machine (LSSVM)) for MP quantification, utilizing six feature extraction methods.

Main Results:

• The SVM model achieved a high F1 score of 0.9817 for rapid MP classification.
• Regression models demonstrated accuracies exceeding 83%, with RF showing the highest overall accuracy.
• The PE-UVE-RF model exhibited excellent performance with R_c^2 of 0.9974 and R_p^2 of 0.9916, validated by hypothesis testing and real sample prediction.

Conclusions:

• THz-TDS combined with machine learning provides a fast, accurate, and effective method for classifying and detecting microplastics in agricultural soils.
• The developed SVM and RF-based models offer significant improvements in speed and accuracy for soil MP analysis.
• This integrated approach holds great potential for environmental monitoring and assessing the impact of microplastics on soil ecosystems.