Dual-channel plasmonic color prints based on deep-learning

When using a metasurface to encode multi-channel color images, in addition to nanostructural shape on the metasurface, efficiently determining nanostructural size based on the coding information has become an urgent problem. Here, we take dual-channel color prints as the research object. By modulating the silicon nanofin’s two arms’ length and the period of the meta-atom, the reflected colors are manipulated under different linearly polarized lights. According to any two color patches, to obtain the nanofin’s appropriate parameters, a dual-task cascaded neural network (DT-CNN) is designed. And the model prediction is compared with the electromagnetic simulation. Then two colorful images are used as the model input. The model output shows that the DT-CNN is useful for dual-channel color prints. In addition, it also can be applied to optical anti-counterfeiting, information storage, color display, and encrypted communication.