【作者】 杨家婧；

【导师】 方群； Jaap M.J.den Toonder；

【作者基本信息】 浙江大学 ， 分析化学， 2022， 博士

【摘要】 近年来,随着花粉过敏人群的增加,人们对于知晓大气中花粉种类与浓度的需求正在逐渐上升。传统的花粉颗粒监测技术依赖长时间的采样和人工目视检测,整套流程耗时且需要专业技术人员参与,难以建立大范围的监测点,普及程度受限。颗粒物光学自动监测技术可通过单颗粒光学信号检测进行实时分析,然而该技术难以结合富集采样技术,对于低浓度的花粉颗粒检测准确度不高。基于静电、气流曳力等原理的捕集技术,结合花粉显微成像技术以及人工智能图像鉴别算法可实现高捕集率、高准确度、高自动化的低浓度花粉实时监测,具有应用于花粉大规模监测的潜力。第一章依据花粉富集和检测方法的不同,对花粉气溶胶监测领域的现状、发展与常见的商业化仪器进行了分类综述。对于先富集再检测的花粉分析技术,详细介绍了花粉被动与主动的富集采样方法、手动和自动检测方法,并分析了每种技术的优劣;对于流式花粉检测技术,主要介绍了利用单颗粒花粉的散射光和激光诱导荧光法实现花粉的鉴别和计数,以实现实时检测。最后,介绍了目前发展较为成熟的基于光学及成像原理的花粉自动识别系统。第二章发展了一种基于静电捕集和显微成像的全集成化花粉实时监测仪。该监测仪集成了无损进样模块、静电捕集模块、自动显微成像模块和自清洁模块,实现了花粉颗粒的定量引入、连续捕集、自动成像、机器视觉图像分析、捕集区域自清洁和循环使用。在三个数量级的大范围花粉浓度内,系统捕集效率均稳定25%左右。此外,我们还提出了用于生成标准浓度的含花粉气体的方法,将花粉无损地传输到捕集模块,以便于仪器性能的定量化评估。与传统的花粉分析仪器相比,本监测仪实现了从捕集、成像到分析全过程的自动化,可实时获取花粉浓度,大大节省人工,提高了分析效率。第三章研制了基于静电-曳力捕集和多通道图像鉴别的自动花粉监测系统。基于空气倍增原理,构建了独立的花粉传输模块,实现了花粉颗粒的无损传输。在静电捕集的基础上增加了气流曳力捕集,将花粉颗粒的捕集效率提升至63%,实现了高采样流量下的高效捕集;设计构建了多通道显微成像模块,可拍摄明场和两个不同波段的荧光通道花粉图片;通过荧光强度比值初步实现了花粉种类的鉴别,同时发展了基于深度学习的Alex Net图像分析算法,区分七种花粉颗粒的准确率达97%以上。本监测仪实现了空气中花粉的无损传输、高效捕集、快速多通道显微成像以及人工智能图像分析的全流程自动化,可快速获取空气花粉信息,为构建大范围的花粉监测网络提供了设备与技术支撑。更多还原

【Abstract】 In recent years,with the increase of pollen allergy cases,the population has an increasing demand for monitoring the types and concentrations of pollen in the air.Traditional pollen particle monitoring relies on long-term sampling combined with manual visual inspection.The whole process is time-consuming and requires the participation of high professional personnel.As a result,it is difficult to establish the pollen monitoring sites in whole word.Optical automatic monitoring instruments can perform real-time analysis through single-particle optical signal detection.However,this type of analysis methods lacks the enriching samples process,which results in low detection accuracy for low-concentration pollen aerosols.Pollen sampling methods based on electrostatic and airflow drag forces combined with pollen microscopic imaging technology and artificial intelligence identification algorithm can realize real-time monitoring of low-concentration pollen with high capture efficiency,high accuracy and high automation.The whole system has the potential to be used for large-scale pollen monitoring around the world.The first chapter starts with pollen enrichment and analysis methods for the monitoring of pollen particles in the air,then summarizes the development and widely used commercial instruments in the field of pollen aerosol monitoring.For the pollen monitoring approaches under the enrichment-analysis mode,the passive and active pollen enrichment sampling methods,manual and automatic analysis techniques are introduced in detail,then the advantages and disadvantages of each technique are analyzed.For the flow-type pollen monitoring approach,the identification and counting of pollen particles are mainly realized by the scattering light and laser-induced fluorescence of single pollen particle,which can realize real-time detection and analysis.Finally,some automatic pollen identification systems based on optics and imaging principles are introduced.The second chapter describes a fully integrated real-time pollen monitor based on electrostatic capture and microscopic imaging.The monitor system integrates nondestructive sampling,electrostatic enrichment,automatic photo analysis,and selfcleaning modules.Through these modules,we have realized quantitative introduction of standard concentration of pollen aerosols,enrichment of pollen grains,real-time automatic detection based on machine vision,self-cleaning and recycling of detection area.In the pollen concentration range in three orders of magnitude,the system capture efficiency is stable at around 25%.The detection module that follows can continuously output microscopic images of captured pollen for subsequent automatic algorithm analysis.Furthermore,we present approaches for generating standard concentrations of particulate gases that can be transferred nondestructively into enrichment modules,which can be applied to quantitatively characterizing instrument performance.Compared with the traditional pollen analyzer,the pollen analyzer we developed realizes the automation of the whole process from collection to analysis.Moreover,it can obtain the pollen concentration in real time,which greatly saves labor costs and improves the analysis efficiency.In chapter three,an automatic pollen monitoring system based on electrostatic-drag force and multi-channel image identification is developed.Based on the principle of air amplification,we have designed and constructed an independent pollen transport module and realized the lossless transport of pollen particles.On the basis of electrostatic pollen capture,airflow drag force capture is added,which allows the pollen capture efficiency increasing to 63%,which realizes the highly efficient capture under high sampling flow.We have also developed a multi-channel microscopic imaging module to obtain the pollen images in bright and fluorescent field.For the pollen automatic identification,we applied the ratio of fluorescence intensity to preliminarily identify pollen species.At the same time,we developed the Alex Net algorithm based on deep learning to distinguish 7 species of pollen and the accuracy is over 97%.The newly developed pollen monitor realizes the whole process automatic of non-destructive transmission,efficient capture,fast multichannel microscopic imaging and artificial intelligence analysis of pollen in the ambient air,which can quickly obtain pollen information,providing technical support for building a large-scale pollen monitoring network.更多还原