Data-Driven Inverse Design of Metasurface for Optical Edge Detection Imaging

Abstract

With the ongoing development of nanophotonic platforms in the field of optical image processing, the forward trial-and-error method is becoming computationally expensive. In this paper, we proposed an efficient inverse design methodology for a nanophotonics platform, which is more computationally effective than a conventional genetic algorithm to get the desired target function. Our methodology includes an RCWA-based genetic algorithm strengthened by a data-driven U-NET model to inversely design the metasurface, enhancing control over the emitted light field for s-polarization selective optical edge detection imaging. The generated metasurface is demonstrated to preserve the strong suppression of background content while retaining only the object's boundaries for edge detection imaging which is crucial for large-scale low power computational imaging, augmented reality, and autonomous driving