Smart polarization and spectroscopic holography for real-time microplastics identification

Microplastics (MPs), tiny plastic particles, pose a significant threat to ecosystems and human health. Ingestion can lead to various health issues, including gastrointestinal problems and death. The production of plastics contributes significantly to greenhouse gas emissions. Current methods for MP detection, such as scanning electron microscopy (SEM) and transmission electron microscopy (TEM), primarily focus on morphological features, which can be unreliable due to environmental weathering. Spectroscopic techniques like Raman microscopy and Fourier transform infrared spectroscopy (FT-IR) offer chemical analysis but suffer from weak signals, long processing times, and complex sample preparation. This necessitates the development of rapid, non-destructive, and non-invasive techniques for accurate MP identification. Polarization imaging (PI) and digital holography (DH) offer potential alternatives. PI captures polarization state changes related to material anisotropy, but manual configuration limits its real-time application. DH captures both amplitude and phase information but individual holographic features are sensitive to environmental changes. This research addresses these limitations by developing a novel method that combines the advantages of PI and DH for improved MP identification and characterization.

Existing methods for microplastic identification have limitations. While techniques like SEM and TEM provide detailed morphological information, they are time-consuming and require sample preparation. Spectroscopic methods such as Raman and FT-IR offer chemical information but often suffer from weak signals and lengthy processing. Polarization imaging (PI) can capture anisotropy and birefringence but usually requires manual adjustment of polarization states, hindering real-time applications. Digital holography (DH) offers high-throughput capabilities but relies on single features that can be affected by environmental variations. Prior work using PI and DH for MP identification has shown limitations in accuracy due to limited feature extraction or sensitivity to environmental changes. This study aims to overcome these limitations by combining polarization, holographic, and textural features for robust MP identification.

The researchers designed a smart polarization and spectroscopic holography (SPLASH) system. The system uses a 532 nm laser, a convex lens for beam collimation, a linear polarizer and quarter-wave plate to control light intensity, and a polarization camera with a Stokes polarization mask (SPM) to simultaneously capture four polarization states (0°, 45°, 90°, and 135°). This setup allows for simultaneous imaging of polarization, holographic, and textural features, effectively incorporating spectroscopic information without a dedicated spectroscopic system. The system captures images with a spatial resolution of 1232 x 1028 pixels, capable of detecting particles as small as 20 µm. Feature extraction involves analyzing texture features (neighborhood gray-tone difference matrix and gray level size zone matrix), Fourier power spectrum (FPS) features (radial and angular summation), holographic features (fringes contrast and transparency), and polarization features (angle of polarization (AoP), degree of linear polarization (DoLP), phase retardation, and optical axis orientation). The extracted features are then used for classification using machine learning algorithms such as ensemble subspace discriminant (ESD) classifier, k-nearest neighbors (KNN) classifier, neural network (NN), and support vector machine (SVM). The performance of the classifiers is evaluated using receiver operating characteristic (ROC) curves and area under the curve (AUC) values. Experiments were conducted using various microplastic materials (PET, PP, PC, PVC, PMMA) and natural particles (young root of plant T.S., *Chlorella*, and *Daphnia magna*).

SPLASH successfully captured distinct polarization, holographic, and texture features for various microplastic materials. Feature correlation analysis revealed strong positive correlations between polarization features and other feature groups, except for DoLP which showed negative correlation with some texture features potentially due to variations in specimen thickness. Texture features alone showed weak discriminative power, highlighting the importance of integrating other feature types. Classification experiments using different machine learning algorithms (ESD, KNN, NN, SVM) demonstrated that polarization features were the most effective for MP identification, achieving AUC values exceeding 0.8 with low variance (<0.05). The combination of polarization, holographic, texture, and FPS features yielded even better performance, achieving an AUC of 0.85 with the ESD classifier. Experiments with natural particles further validated the system's ability to discriminate between MPs and natural materials. The AoP plots provided clear visual distinctions among different materials, further supporting the findings.

The results demonstrate the effectiveness of the SPLASH system in identifying microplastics. The integration of polarization, holographic, and textural features, combined with machine learning algorithms, significantly improves the accuracy and efficiency of MP identification compared to existing methods. The ability to simultaneously capture multi-dimensional features eliminates the need for separate spectroscopic systems, simplifying the process and making it more suitable for real-time applications. The high accuracy and low variance achieved in the classification experiments indicate the robustness of the method. The successful discrimination between MPs and natural particles highlights the potential of SPLASH for in situ MP detection in complex environmental samples. Future work could focus on improving image quality in turbid water environments and developing more compact, portable devices for field use.

The SPLASH system presents a novel, accurate, and efficient method for real-time microplastic identification. The integration of polarization, holographic, and textural features, combined with machine learning, significantly improves the accuracy and efficiency compared to existing techniques. The system’s ability to discriminate MPs from natural particles and the high classification accuracy demonstrate its potential for environmental monitoring and pollution assessment. Future research directions include developing high-throughput microfluidic systems and compact, portable devices for broader applications in diverse environments.

While the SPLASH system shows high accuracy, further improvements could be made. The performance might be affected in highly turbid water environments due to light scattering. The current system is laboratory-based; developing a portable device for field applications would enhance its practicality. Extending the dataset to include a wider range of microplastic types and sizes would further improve the generalizability of the findings. Finally, more thorough investigation of the effects of environmental weathering and aging on the measured features would be beneficial.