Home | Research | Teaching | Publications | Book | Service

Research Interests

• Electromagnetic inverse scattering and computational imaging

• Physics-inspired machine learning for solving engineering problems

• Millimeter wave SAR imaging

• Optical encryption and authentication

• Microwave impedance scanning microscopy

• Optical and infrared scanning microscopy

• Metamaterials

Google Scholar: https://scholar.google.com.sg/citations?user=Z-GXeiwAAAAJ&hl=en&oi=ao

Introduction to Research Topics:

My main research areas are electromagnetic inverse problems, primarily inverse scattering and related computational imaging. This field is highly inter-disciplinary, drawing upon knowledge from wave physics, data analytics (mainly optimization and low-rank representation), scientific computing, and signal processing.

Introduction to inverse problems: An inverse problem in science is the process of calculating from a set of observations/measurements the causal factors that produce them. It is called an inverse problem because it starts with the results and then calculates the causes. This is the inverse of a forward problem, which starts with the causes and then calculates the results. For example, for the forward problem that "if we know the structure of the inner organs of a patient, what kind of X-ray images would we obtain?" the corresponding inverse problem is the well-known tomography problem, "given a set of X-ray images of a patient, what would be the three-dimensional structure of his inner organs?"

Electromagnetic inverse problems: Electromagnetic energy is used to probe an unknown target and then the response of the target to electromagnetic excitation is measured. It is often the case that the image of the target is not easy to obtain due to either the far distance of the target from sensors or the presence of barrier between it and sensors. This kind of imaging problem can often be tackled by solving electromagnetic inverse problems. Thanks to wide electromagnetic spectrum, probing using electromagnetic energy provides a variety of imaging capacities, in terms of resolution, contrast, penetration, cost, safety, which enables end-users to choose the desired frequency range for their specific applications.

Real-world applications: Electromagnetic inverse problems have been of great interest to researchers due to their importance in industrial, military, medical, geophysical, and civil applications. To name a few, through-the-wall imaging, screening of breast cancer, material characterization, detection of concealed weapon at airport, non-destructive evaluation of products, remote sensing, detection of subsurface objects, exploration of mine and oil, are all related to electromagnetic inverse problems.

Research type: My research on electromagnetic inverse problems focuses on theory, modeling, algorithm, and computation. To numerically reconstruct unknown targets can be generally considered as a parameter-estimation problem, which needs many data-analytics techniques, such as optimization, machine learning, compressive sensing, low-rank representation, and regularization. However, a major difference from a general data-analytics problem is that the data collected in inverse problems are often governed by physical laws. Keeping in mind that the research into the inverse problem requires a deep or fairly good understanding of the corresponding forward problem, I always hesitate to directly apply a general optimization method to a high-dimensional nonlinear problem, where the original forward problem is iteratively evaluated. I am convinced that insights and intuitions, no matter whether they are mathematical, physical, or engineering, potentially help us to solve the problem in a more efficient and elegant way.

Recent work on machine learning: My team has recently applied machine learning (ML) to solving two types of engineering problems (ISP and EIT). My approach of using ML emphasizes on drawing insights from physics, which helps to avoid a brute-force (or black-box) way of using ML. In many real-world applications, data are collected by sensors (optical, acoustical, microwave wave, electric current/voltage, heat….), automatically governed by physical laws. Some of these physical laws present well-known mathematical properties (even analytical formulas), which do not need to be learnt by training with a lot of data. It is important to answer the question of how best to combine ML with knowledge of the underlying physics as well as traditional non-ML techniques.