基于红外成像光谱的微塑料人工智能识别机理与定量分析方法研究

Micro plastics is a new kind of contaminants in marine food and environment. Accurate component identification and quantitative determination of micro plastic are essentially important for research and control of micro plastic pollution. Spectroscopic technique has potential in the detection of micro plastic. The main barrier of its effective application is the difficulty of spectral resolution and quantitative determination. With the support of microscopic Fourier transform infrared (FTIR) imaging, this project aims to realize the identification of micro plastics through the resolution of infrared spectra by machine learning algorithms, and the quantitative determination of micro plastic through the autonomous counting by machine vision. Through the study on the relationship between FTIR spectral images and components, size, and weathered levels of micro plastic, the fitting models of the infrared spectra of micro plastics will be established. The autonomous algorithms of component identification will be developed to identify different types of micro plastics. An image fusion algorithm using multiple featured spectral images will be developed to combine multiple features of spectral images into one image. Regional statistical algorithms of fused image will be developed to determine the quantity of micro plastic. The accuracy of developed algorithm and models will be verified and optimized by comparing to the data of micro plastic from kinds of sea food. This project can provide a method for the effective identification and quantification of micro plastic.

微塑料是一种新型海洋环境与食品污染物，其鉴别和丰度分析是海洋微塑料污染研究与控制的基础。光谱分析技术是目前微塑料检测的一种可行手段之一，但目前最主要的问题是光谱数据解析困难，且难以定量检测。因此，我们提出基于显微傅里叶变换红外光谱(FTIR)成像技术，利用机器学习算法自动解析红外光谱，以准确识别不同微塑料，并利用机器视觉技术自动统计其含量。本项目将明确不同组分、不同尺度、不同风化程度微塑料的红外光谱特征和不同波长红外图像特征，建立微塑料红外光谱特征的解析模型，提出组分自主识别算法，实现不同微塑料的准确识别；建立微塑料多特征波长红外图像的融合模型，实现不同微塑料空间分辨；建立融合微塑料空间分辨图像的区域统计算法，实现不同微塑料定量分析；通过实际微塑料污染物的FTIR显微成像实验，验证模型算法的识别准确性，为微塑料污染物的快速、自主识别和定量分析检测提供一种有效手段。

微塑料是一种新型海洋环境与食品污染物，其鉴别和丰度分析是海洋微塑料污染研究与控制的基础。光谱分析技术是目前微塑料检测的一种可行手段之一，但目前最主要的问题是光谱数据解析困难，且难以定量检测。我们基于显微高光谱成像技术，利用机器学习算法自动解析光谱和图像，准确识别了不同微塑料，并利用机器视觉技术自动统计了其含量。明确了不同组分、不同尺度微塑料的光谱特征和不同波长图像特征，提出了组分自主识别算法，实现了不同微塑料的空间分辨；建立了微塑料空间分辨图像的区域统计算法，实现了微塑料定量分析；基于模糊图像实现了高光谱图像的复原，为微塑料污染物的快速、自主识别和定量分析检测提供一种有效手段。