Establishing Confidence in Wavefield Images for Agricultural and Biomedical Applications

点击翻译按钮获取中文摘要

Wavefield imaging uses electromagnetic or acoustic wave interrogation to non-invasively reconstruct the constitutive structure of a target (i.e., "see inside"). A wavefield imaging system can be divided into two parts: the hardware required to interrogate a target and collect data, and one of many possible imaging algorithms used to convert the data into an image of the target. My research program focuses on developing and applying novel and improved wavefield imaging algorithms to two application areas: breast cancer detection and monitoring, and grain storage monitoring. While these applications require different hardware systems, they are solved using the same algorithms. Wavefield imaging is a very challenging problem, and the quality of an image is both target and algorithm-parameter dependent. Validation of imaging systems has historically relied on comparing images to the true target. In practice the true target is unknown, and validating an algorithm for some targets does not imply that it will produce accurate images of all potential targets. Fortunately, imaging algorithms have a number of tunable parameters (e.g., frequency of interrogation, assumed prior information), that can be used to produce multiple images of the same target. If these images are similar, confidence grows; if they differ, confidence wanes. This research program addresses establishing confidence in wavefield images. I will create a framework in which wavefield imaging users can generate both a wavefield image and an associated image confidence map. Confidence will be established by efficiently generating a parameter-diverse set of target images and applying image fusion/comparison and image sensitivity analysis. This framework squarely addresses the critical missing step of image validation needed for wavefield imaging users to adopt the technology as a decision-directing tool. Clinically robust and dependable microwave breast cancer imaging will provide a low-cost, low-risk, minimally invasive mass-screening tool. Radio-frequency grain storage monitoring is a Canadian-pioneered potentially disruptive technology that is poised to become a global standard for securing our stored grain. The program will train nine HQP in the areas of applied mathematics, computations, wave physics, inverse problems, machine learning and optimization. These HQP, equipped with essential skills in wavefield imaging, are ideal to fill a skills gap that exists in this growing Canadian industry.