A large amount of nanomaterial characterization data has been routinely collected by using electron microscopes and stored in image or video formats. A bottleneck in making effective use of the image/video data is the lack of the development of sophisticated data science methods capable of unlocking valuable material pertinent information buried in the raw data. To address this problem, the research of this dissertation begins with understanding the physical mechanisms behind the concerned process to determine why the generic methods fall short. Afterwards, it designs and improves image processing and statistical modeling tools to address the practical challenges. Specifically, this dissertation consists of two main tasks: extracting useful information from images or videos of nanomaterials captured by electron microscopes, and designing analytical methods for modeling/monitoring the dynamic growth of nanoparticles. In the first task, a two-pipeline framework is proposed to fuse two kinds of image information for nanoscale object detection that can accurately identify and measure nanoparticles in transmission electron microscope (TEM) images of high noise and low contrast. To handle the second task of analyzing nanoparticle growth, this dissertation develops dynamic nonparametric models for time-varying probability density functions (PDFs) estimation. Unlike simple statistics, a PDF contains fuller information about the nanoscale objects of interests. Characterizing the dynamic changes of the PDF as the nanoparticles grow into different sizes and morph into different shapes, the proposed nonparametric methods are capable of analyzing an in situ TEM video to delineate growth stages in a retrospective analysis, or tracking the nanoparticle growth process in a prospective analysis. The resulting analytic methods have applications in areas beyond the nanoparticle growth process such as the image-based process control tasks in additive manufacturing.
A large amount of nanomaterial characterization data has been routinely collected by using electron microscopes and stored in image or video formats. A bottleneck in making effective use of the image/video data is the lack of the development of sophisticated data science methods capable of unlocking valuable material pertinent information buried in the raw data. To address this problem, the research of this dissertation begins with understanding the physical mechanisms behind the concerned process to determine why the generic methods fall short. Afterwards, it designs and improves image processing and statistical modeling tools to address the practical challenges. Specifically, this dissertation consists of two main tasks: extracting useful information from images or videos of nanomaterials captured by electron microscopes, and designing analytical methods for modeling/monitoring the dynamic growth of nanoparticles. In the first task, a two-pipeline framework is proposed to fuse two kinds of image information for nanoscale object detection that can accurately identify and measure nanoparticles in transmission electron microscope (TEM) images of high noise and low contrast. To handle the second task of analyzing nanoparticle growth, this dissertation develops dynamic nonparametric models for time-varying probability density functions (PDFs) estimation. Unlike simple statistics, a PDF contains fuller information about the nanoscale objects of interests. Characterizing the dynamic changes of the PDF as the nanoparticles grow into different sizes and morph into different shapes, the proposed nonparametric methods are capable of analyzing an in situ TEM video to delineate growth stages in a retrospective analysis, or tracking the nanoparticle growth process in a prospective analysis. The resulting analytic methods have applications in areas beyond the nanoparticle growth process such as the image-based process control tasks in additive manufacturing.
