RSUSSH 2020
Surface Plasmon Phase Retrieval using Deep Learning
Dr. Suejit Pechprasarn
Invited speaker G1-1
Abstract
Surface Plasmon Resonance (SPR) has been a gold standard for biosensing. The surface plasmons are a confined electromagnetic wave phenomena propagating on surface of noble metals. The surface plasmons are sensitive to its surrounding medium and can be employed in sensing applications. It has been very well established that measuring the phase response of the surface plasmons is more sensitive and more robust compared to intensity. To measure the phase, of course, an interferometer is required. This is burden to the complexity to the optical alignment. It also requires a well-controlled experimental condition, such as, vibration isolation system. Here, we propose a novel approach to perform surface plasmon phase retrieval using pattern recognition though deep learning. We demonstrate the feasibility of using the deep learning to recover real and imaginary parts of simulated back focal plane image. The accuracy of the trained network is validated with experimental back focal plane.
QUESTIONS & ANSWERS
Thank you very much for your presentation Dr. Suejit. I wonder if there are potentially issues with training a neural network using simulated data? Specifically, that the network might model the function used to simulate the data, but then not generalize well to real-world data (e.g. from a different sample/optical system). Have you tested this hypothesis?
@30 Apr 20, 12:16 PM That's very good point. If the simulated data do not represent the real-world data well enough, this method would not work. Here the network is trained for the SPR sample only and with the specific objective lens. If the sample or the objective lens setting were changed to some other values. This deep learning would not work. In this work, I would like to point out that for a well defined csse, it is possible to use a simulated data to train the network and the network can still perform reasonably well with a real experimental result under a well defined condition.
@30 Apr 20, 01:32 PM Thank you. It will be interesting to see how this technology can continue to be of use in optical and bio-sensing applications
@30 Apr 20, 04:25 PM