Department of Chemistry, 6640Idaho State University, Pocatello, USA.
Appl Spectrosc. 2020 Sep;74(9):1167-1183. doi: 10.1177/0003702820923993. Epub 2020 Jun 1.
Microplastic research is an emerging field. Consistent accurate identification of microplastic polymer composition is vital for understanding the effect of microplastic pollution in the environment. Fourier transform infrared (FT-IR) spectroscopy is becoming commonplace for identifying microplastics. Conventional spectral identification is based on library searching, a process that utilizes a search algorithm against digital databases containing single spectra of pristine reference plastics. Several conditions on environmental microplastic particles such as weathering, additives, and residues cause spectral alterations relative to pristine reference library spectra. Thus, library searching is vulnerable to misidentification of microplastic samples. While a classification process (classifier) based on a collection of spectra can alleviate misidentification problems, optimization of each classifier (tuning parameter) is required. Additionally, erratic results relative to the particular optimized tuning parameter can occur when microplastic samples originate from new environmental or biological conditions than those defining the class. Presented in this study is a process that utilizes spectroscopic measurements in a hybrid fusion algorithm that depending on the user preference, simultaneously combines high-level fusion with low- and mid-level fusion based on an ensemble of non-optimized classifiers to assign microplastic samples into specific plastic categories (classes). The approach is demonstrated with 17 classifiers using FT-IR for binary classification of polyethylene terephthalate (PET) and high-density polyethylene (HDPE) microplastic samples from environmental sources. Other microplastic types are evaluated for non-class PET and HDPE membership. Results show that high accuracy, sensitivity, and specificity are obtained thereby reducing the risk of misidentifying microplastics.
微塑料研究是一个新兴领域。为了了解环境中微塑料污染的影响，始终如一地准确识别微塑料的聚合物组成至关重要。傅里叶变换红外（FT-IR）光谱分析已成为识别微塑料的常用方法。传统的光谱识别基于库搜索，该过程利用搜索算法对包含原始参考塑料单光谱的数字数据库进行搜索。环境微塑料颗粒的一些条件，如风化、添加剂和残留物，会导致光谱相对于原始参考库光谱发生变化。因此，库搜索容易导致微塑料样品的误识别。虽然基于一系列光谱的分类过程（分类器）可以减轻误识别问题，但需要对每个分类器（调谐参数）进行优化。此外，当微塑料样品来自新的环境或生物条件，而不是定义类别的条件时，相对于特定优化的调谐参数，可能会出现不稳定的结果。本研究提出了一种利用光谱测量的混合融合算法的过程，该算法根据用户的偏好，同时根据非优化分类器的集合，结合高低层次融合，将微塑料样品分配到特定的塑料类别（类）中。该方法使用 FT-IR 对来自环境源的聚对苯二甲酸乙二醇酯（PET）和高密度聚乙烯（HDPE）微塑料样品进行二进制分类，用 17 个分类器进行了演示。还评估了其他微塑料类型是否属于非类 PET 和 HDPE。结果表明，该方法获得了较高的准确性、灵敏度和特异性，从而降低了微塑料误识别的风险。
