Fig. 1 | Nature Communications

Fig. 1

From: Machine-learning reprogrammable metasurface imager

Fig. 1

Working principle of the real-time digital-metasurface imager. a The proposed machine-learning metasurface imager can be optimized for different kinds of scenes. The optimization is performed by training the manifold representing the metasurface configuration (the surfaces) to be close to the training data (scatters). b The illustration of training the reprogrammable imager by using the PCA method (see Supplementary Note 2), in which the training person is Hengxin Ruan (coauthor). c The map of coding metasurface and the unit cell (see Supplementary Fig. 1, 2). d The illustration of real-time reprogrammable metasurface imager imaging a moving person behind a wall. Two reconstructions are displayed at right (more experimental results in Supplementary Videos 3, 4 and 5)

Back to article page