Classification and Quantification of Microplastics (<100 μm) Using a Focal Plane Array–Fourier Transform Infrared Imaging System and Machine Learning

Abstract

Microplastics are defined as microscopic plastic particles in the range from few µm and up to 5 mm. These small particles are classified as primary microplastic when they are manufactured in this size range, whereas secondary microplastics arise from the fragmentation of larger objects. Microplastics are a widespread emerging pollutant and investigations are underway to determine potential harmfulness to biota and human health. However, progress is hindered by the lack of suitable analytical methods for rapid, routine and unbiased measurements. This work aims to develop an automated analytical method for the characterization of small microplastic (< 100 µm) using micro Fourier Transform Infrared (µ-FTIR) hyperspectral imaging and machine learning tools. Partial least squares discriminant analysis (PLS-DA) and soft independent modelling of class analogy (SIMCA) models were evaluated, applying different data pre-processing strategies for classification of nine of the most common polymers produced worldwide. The hyperspectral images were also analyzed to quantify particle abundance and size automatically. PLS-DA presented a better analytical performance in comparison with SIMCA models with higher sensitivity, sensibility and lower misclassification error. PLS-DA was less sensitive to edge effects on spectra and poorly focused regions of particles. The approach was tested on a seabed sediment sample (Roskilde Fjord, Denmark) to demonstrate the method efficiency. The proposed method offers an efficient automated approach for microplastic polymer characterization, abundance numeration and size distribution with substantial benefits for methods standardization.

A Study of Microplastic Particles in Danish Tap Water

“…Furthermore, multivariate models are faster than library searching with results obtained in few minutes and, can in addition, can reduce bias in MP identification by the implementation of statistical evaluation of the models. Some methods have already been developed to apply this approach for MP analysis [28,29], and they could be a future alternative to retrieve MP information from spectroscopy data obtained in this field.…”

Automated analysis of microplastics based on vibrational spectroscopy: Are we measuring the same metrics?

“…However, the visualization of microplastics based on the integration of a single peak cannot cope with different polymer types, and the considerable amount of data makes this approach unable to be widely used in the automated analysis of microplastics 31 . Within the past two years, multivariate statistics, data mining, and machine learning methods, such as principal component analysis 32 , cluster analysis 32 , random decision forest (RDF) classifiers 33 , and partial least squares discriminant analysis 17 , were applied to process hyperspectral imaging. These methods make it possible to reduce the dimensionality of hyperspectral infrared imaging and detect and quantify microplastics.…”

“…Glass fiber membranes and polymer-type membranes are not applicable due to their highly scattering surface or their strong infrared absorption characteristics. The aluminum oxide filter has been widely used 2,17,[59][60][61] because it has no infrared absorption above 1250 cm -1 59 . The authors recommended FPA-FTIR imaging with FPA to measure the transmission spectra of particles loaded on the aluminum oxide (Fig.…”

“…In contrast, the automated analysis of microplastics based on vibrational spectroscopy (both Raman and infrared) is more compatible with existing studies and sample pretreatment methods. There exist three strategies for the automated analysis of microplastics: laser direct infrared (LDIR)-based particle analysis 15 , Raman-based particle analysis 16 , and focal plane array (FPA)-FTIR imaging 2,[17][18][19] . They are based on different instruments and principles.…”

Automated analysis of microplastics based on vibrational spectroscopy: Are we measuring the same metrics?

“…Glass fiber membranes and polymer-type membranes are not applicable. The aluminum oxide filter has been widely used 2,17,[51][52][53] because it has no infrared absorption above 1250 cm -1 51 . The recommend FPA-FTIR imaging uses FPA to measure the transmission spectra of particles loaded on the aluminum oxide (Fig.…”