Application and exploration of hyperspectral imaging technology in soil microplastic detection

The problem of soil microplastic pollution is becoming increasingly serious, but in-depth research on microplastics in terrestrial ecosystems is still insufficient. This study aims to directly identify and visualize the distribution of microplastics in soil by combining hyperspectral imaging technology with advanced chemometric algorithms, thereby filling this research gap.

1. Sample collection and processing

First, we found and collected some weathered plastic fragments and 5 cm thick soil samples around them on the soil surface, with a total weight of about 3 kg. After these samples were brought back to the laboratory, they were divided into two groups. One group was used to extract and identify the specific components of microplastics by saturated NaCl aqueous solution method; the other group was used to build a microplastic identification model based on hyperspectral imaging technology and chemometric algorithms.

To simulate the existence of microplastics in real soil environments, we further prepared simulated soil samples. Through manual shearing and sieving, we divided the extracted microplastics (white and black) into two particle size ranges of 1-5 mm and 0.5-1 mm, and mixed them with natural materials such as fresh leaves, withered leaves, rocks and branches to simulate the complex field soil environment. All soil samples were dried in a vacuum oven at 80°C for 8 hours to remove moisture and ensure the accuracy of the experiment.

2. Hyperspectral Image Acquisition and Data Analysis

Using a hyperspectral imaging system, we conducted a comprehensive scan of simulated soil microplastic samples and obtained hyperspectral images containing rich spectral information. In the images, different materials (such as white microplastics, black microplastics, fresh leaves, etc.) are marked with different colors to facilitate subsequent analysis.

By analyzing the spectral curves of the region of interest (ROI) of each material on the image, we found that fresh leaves, rich in chlorophyll, exhibit significant spectral characteristics in the visible light region, making them easy to distinguish from other materials. In contrast, white and black PE microplastics differ in spectral characteristics, especially black PE microplastics, which have the lowest reflectivity in the entire spectral range, making identification more difficult.

3. Comparison and Optimization of Supervised Classification Methods

In order to find the best algorithm for microplastic identification, we used three supervised classification methods: multivariate discriminant analysis (MD), machine learning (ML), and support vector machine (SVM). By calculating the precision (P) and recovery rate (R) of each method, we found that the SVM algorithm showed a higher signal-to-noise ratio and less background noise when processing hyperspectral images, thus significantly improving the identification of microplastics.

We conducted classification tests on microplastics of different particle sizes (1-5 mm and 0.5-1 mm). The results showed that for microplastics with larger particle sizes, the SVM algorithm was able to achieve higher recognition accuracy; and for microplastics with smaller particle sizes, the recognition effect was significantly improved by optimizing image morphological preprocessing (such as erosion and dilation operations).

IV. Model Verification and Extended Application

To verify the wide applicability of the model, we collected six household plastic polymers of different colors and chemical compositions and tested their recognition effects under hyperspectral imaging technology. The results showed that for six common microplastics with particle sizes of 1-5 mm and 0.5-1 mm, the model showed good recognition ability, with average accuracy and recovery rates reaching high levels. In particular, the recognition effect of colored microplastics was particularly outstanding due to their more obvious spectral characteristics.

V. Summary and Outlook

This study successfully combined hyperspectral imaging technology with chemometric algorithms to achieve direct identification and visualization of microplastics in soil. By comparing different supervised classification methods, we found that the SVM algorithm has significant advantages in microplastic identification. In addition, the study also revealed the impact of microplastic particle size on identification and proposed corresponding optimization strategies.

In the future, we plan to further expand the application scope of this technology, such as exploring the impact of different soil types and environmental conditions on microplastic identification, and developing more portable and efficient hyperspectral imaging equipment to meet the needs of rapid on-site detection. At the same time, we will continue to optimize the algorithm model, improve the recognition accuracy and stability, and provide more powerful technical support for the monitoring and control of soil microplastic pollution.