复杂海岸环境微塑料可见-红外波谱的多源开集识别机制研究

Microplastics pollution in marine and coastal environments is an area of growing concern, it plays a key role in the world's marine litter problem.The environments plastics sample display in large diversity and has great abundance in China coastal. The possibility of coupling Fourier Transform Infrared Spectroscopy (FTIR) with microscopy (micro-FTIR) enables the noninvasive and non-destructive detection of icroplastics. Hence we propose a Multiple Classifier System (MCS) which is composed of different types of One-Class Classifier (OCC) to do the classification, and a new robust regression method is used to increase the ability to interpret the spectra of environmental samples. The aging model of environmental samples is established based on the fusion of both the spectral and the image information, and the parameter as carbonyl groups, color, textural are introduced in this model. This research can do identification of microplastics （>1mm）in coastal environments with high precision .Microplastics management is one of the most important needed tasks in order to save the world environment. This project will support the development of microplastics pollution regulation and contribute to the marine pollution treatment.

微塑料是一种近年来备受关注的新型海洋污染物，在我国沿海地区的微塑料污染具有负荷高、丰度大、类型复杂等特点。本项目利用显微红外技术快速无损地对海岸环境收集的微塑料样本进行图谱信息采集，为样本老化行为的阶段性检测提供了保障。项目提出采用单类集成多类的开集识别模型，配合自适应最小残差判决原则对复杂基质下多样性微塑料样本的进行精确识别。结合红外光谱和微区图像信息及引入裂纹密度、泛黄色度和羰基比例等参数，建立基于微塑料内部结构和外部形态的多源信息融合模型揭示海岸环境样本老化行为规律。本项目可以实现复杂海岸环境微塑料样本（>1mm）的无损快速精准解析，为我国海岸环境微塑料污染监管提供科学支撑，对微塑料污染治理具有重要意义。

环境微塑料样本的多样特性和其自身携带干扰信息经常为检测带来很大障碍。项目采用单类集成多类的模型对复杂海岸环境的多样性微塑料样本进行开集识别，该多类开集识别模型具有开放性能够实时进行更新，对环境样本的识别性精度高达97.4%，利用基于自适应最小残差判决原则的建模的分类器进一步提升了环境样本的识别精度。研究中发现采用单一的红外光谱特征或者单一样本外部形态特征，难以满足复杂环境背景下对微塑料样本老化行为进行精准解析的需要。项目采用了光谱-图像融合模型对环境微塑料样品的不同降解程度进行有效解析，准确率可达97.1%。解析能力较单变量模型相比有了显著提高，该方法不仅提高了模型的鲁棒性，还更有效地结合了环境样本的多样性特性。该项目为我国海岸环境微塑料污染的监管治理提供科学依据。