Figure 12.

Principle of the Machine-Learning-Driven Microwave Imager and Experimental Results

(A) The proposed machine-learning metamaterial imager is optimized by the training samples. The scene x is compressed through the matrix H: y = Hx. By minimizing the difference between the reconstructed scenes and the original scenes, the optimized matrix H is determined.

(B) Photo of a 2-bit programmable metamaterial.

(C) The sketch of a meta-atom.

(D) Four cases of experimental tests.

(E and F) The experimental results by using the developed imager trained with PCA and random projection, respectively. Reference: L. Li, NC (2019)