Longitudinal Modelling and Sequential Monitoring of Image Data Streams

在与地球和环境监测、制造业、医学研究和许多其他相关的成像应用中，所收集的图像数据通常是数据流的形式，因为新图像是随时间顺序获取的。在这样的应用中，一个基本的任务是监视图像序列，以查看所观察到的图像的基本纵向过程是否随时间显著变化。该项目旨在开发新颖有效的统计方法来回答这个问题。由于图像序列监测的广泛应用，该项目将通过其在不同学科和领域的应用产生更广泛的社会影响。开源R包将免费开发和分发，以方便从业者使用。还将开发一个门户网站，供研究人员个人试用所提出的方法。PI计划将研究成果融入教育活动，包括开发新的课程模块，指导博士生，此外，我们亦会为本地中学生举办课外活动，以提高他们对数据建模及科学研究的兴趣，并为科学、技术、工程及数学方面的人才发展作出贡献。本计画旨在发展一种弹性的纵向模式化方法与一种有效的序列监控机制，以分析影像资料流，并研究其统计特性。所提出的描述给定时间间隔内观测图像的纵向模型是灵活的，并且其估计过程在去除噪声的同时具有边缘保持特性。该方法既能适应观测图像间的几何错位，又能适应观测图像数据的时空相关性。所提出的图像监测方法可以考虑所观察到的图像数据流的动态纵向图案。为此，将在图像监测之前适当地执行图像预处理，包括图像去噪和图像配准。该奖项反映了NSF的法定使命，并已被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估的支持。

In imaging applications related to earth and environmental monitoring, manufacturing industries, medical studies and many others, collected image data are often in the form of data streams in the sense that new images are acquired sequentially over time. In such applications, one fundamental task is to monitor the image sequence to see whether the underlying longitudinal process of the observed images changes significantly over time. This project aims to develop novel and effective statistical methods for answering this question. Because of the wide applications of image sequence monitoring, this project will have broader impacts on society through its applications in different disciplines and areas. Open source R packages will be developed and distributed freely for convenient use by practitioners. A web portal will also be developed for individual researchers to try the proposed methods. The PI plans to integrate the research results into educational activities, including the development of new curriculum modules, the mentoring of Ph.D. students, and outreach to local high schools students for after-school activities to raise their interests in data modeling and scientific research, and contribute to the workforce development in Science, Technology, Engineering and Mathematics. This project aims to develop a flexible longitudinal modeling approach and an effective sequential monitoring scheme for analyzing image data streams, and study their statistical properties. The proposed longitudinal model for describing observed images in a given time interval is flexible, and its estimation procedure has the edge-preservation property while removing noise. It can accommodate both geometric misalignments among observed images and spatio-temporal data correlation in the observed image data. The proposed image monitoring approach can account for dynamic longitudinal patterns of the observed image data streams. To this end, image pre-processing, including image denoising and image registration, will be performed properly before image monitoring. The proposed methods will consider both cases where the observation times are equally or unequally spaced.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.