Optical analog cross-correlation computing for pattern recognition based on a dielectric metasurface

Optical analog computing has been demonstrated to break the limitations of digital computation and raises a higher level of computational speed, throughput, and efficiency. The employment of optical analog computing for pattern recognition facilitates expeditious segmentation and highlighting of the target pattern from the background, to improve computing efficiency. The scheme of optical analog cross-correlation computing implemented via a dielectric metasurface has been experimentally proven to enable real-time localization and recognition of the specified target pattern at different positions. Specifically, by approximating the Fresnel diffraction process as a Fourier transform, a compact, flexible, and precise optical analog computing kernel is designed, further enhancing system integration. The kernel can be applicable to other functions of optical analog computing as well. The proposed scheme promotes the development and application of optical analog computing in optical information processing and computer vision fields, such as bio-detection and information retrieval.