Janus meta-imager: asymmetric image transmission and transformation enabled by diffractive neural networks

The asymmetric imaging device is a crucial and highly desired component in optical and electromagnetic systems. However, most existing asymmetric imaging devices are based on active or nonlinear materials and are limited to one-directional applications. This paper reports a method to realize asymmetric image transmission and transformation in two opposite directions, respectively, based on diffractive deep neural networks (D²NNs), named Janus meta-imager. It is a passive device composed of several diffractive layers that are well-trained using deep-learning-based algorithms. We first experimentally fabricate and validate this Janus meta-imager in the near-infrared (NIR) band, which agrees well with simulation results, thus verifying the asymmetric imaging function. This scheme has the merits of high-speed all-optical processing, low energy consumption, and small size, offering potential applications in all-optical encryption and information storage.