Neural network-based information metasurface microwave imager

Microwave imaging, a non-contact identification technology, has garnered widespread research attention for its high privacy protection and environmental adaptability. However, existing microwave imaging technologies are limited by expensive equipment and complex algorithms, making them unsuitable for large-scale deployment. We propose an information metasurface microwave imaging system based on a neural network, which features low cost, low complexity, and high efficiency. By employing beamforming function on a compact, dual-band, and easily manufacturable 1-bit information metasurface, precise imaging of large-scale targets is facilitated, thereby reducing the hardware burden. Moreover, a three-layer convolutional neural network is utilized to reconstruct imaging results. The training occupies minimal computational resources, and the model converges rapidly. Ultimately, PSNRs of the imaging result are mainly concentrated between 60 and 70 dB. This imager showcases remarkable performance in indoor imaging and small robot behavior recognition tasks, holding promise in future Internet of Things (IoT) monitoring and smart home deployments.