A Critical Review on Applying Machine Learning to the Identification of Atmospheric Microplastics and Nanoplastics
Is this project an undergraduate, graduate, or faculty project?
Undergraduate
Project Type
individual
Campus
Daytona Beach
Authors' Class Standing
Simon Cole, Junior
Lead Presenter's Name
Simon Cole
Lead Presenter's College
DB College of Engineering
Faculty Mentor Name
Marwa El-Sayed
Abstract
The pervasive distribution of micro- and nanoplastics (MPs/NPs) across various ecosystems and environmental compartments pose a significant yet not fully quantified threat to environmental and human health. This risk necessitates the prompt development of powerful and efficient detection methodologies. Traditional MPs/NPs imaging and detection methods provide limited adequacy due to constraints in power, accuracy, and analysis speed. The use of machine learning (ML) algorithms can be utilized in combination with these methods to facilitate improved accuracy and automated counting. The aim of this study is to perform a methodical analysis of the literature surrounding the application of ML techniques to the identification of MPs/NPs with an emphasis on identifying particles of sizes small enough to remain suspended and/or dispersed in the atmosphere. While the abundance of these plastic particles in marine environments has been extensively researched for decades, their prevalence in the atmosphere has only recently emerged as a significant research question, making it comparatively understudied. In order for an ML algorithm to identify MPs/NPs of a size less than 0.5mm, the data must first be generated by a chemical imaging method. Herein, we examine several imaging methodologies including optical, Fourier Transform Infrared Spectroscopy (FTIR), Raman Spectroscopy, Digital Holography (DH), Near-Infrared Spectroscopy (NIR), and Hyperspectral imaging. For each imaging method, we first define a size range at which it is effective, then systematically review the advantages and limitations of different ML algorithm options. Implications for this work showcase the effectiveness of ML in combination with traditional imaging techniques to accurately identify MPs/NPs.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
No
A Critical Review on Applying Machine Learning to the Identification of Atmospheric Microplastics and Nanoplastics
The pervasive distribution of micro- and nanoplastics (MPs/NPs) across various ecosystems and environmental compartments pose a significant yet not fully quantified threat to environmental and human health. This risk necessitates the prompt development of powerful and efficient detection methodologies. Traditional MPs/NPs imaging and detection methods provide limited adequacy due to constraints in power, accuracy, and analysis speed. The use of machine learning (ML) algorithms can be utilized in combination with these methods to facilitate improved accuracy and automated counting. The aim of this study is to perform a methodical analysis of the literature surrounding the application of ML techniques to the identification of MPs/NPs with an emphasis on identifying particles of sizes small enough to remain suspended and/or dispersed in the atmosphere. While the abundance of these plastic particles in marine environments has been extensively researched for decades, their prevalence in the atmosphere has only recently emerged as a significant research question, making it comparatively understudied. In order for an ML algorithm to identify MPs/NPs of a size less than 0.5mm, the data must first be generated by a chemical imaging method. Herein, we examine several imaging methodologies including optical, Fourier Transform Infrared Spectroscopy (FTIR), Raman Spectroscopy, Digital Holography (DH), Near-Infrared Spectroscopy (NIR), and Hyperspectral imaging. For each imaging method, we first define a size range at which it is effective, then systematically review the advantages and limitations of different ML algorithm options. Implications for this work showcase the effectiveness of ML in combination with traditional imaging techniques to accurately identify MPs/NPs.