Meta-optic Image Differentiation for High-speed Target Recognition

There is a growing need for faster target recognition algorithms that also reduce power consumption and information bandwidth. Optical analog computing systems can fill this need by offloading the most computationally expensive tasks to the optics; however, traditional systems are too heavy and bulky for size, weight, and power-constrained platforms. In this work, we will develop a mid-wave infrared meta-optic that performs image differentiation operations in a flat, compact form factor. Meta-optics use subwavelength elements to impart local phase shifts and can achieve functionalities beyond that of refractive optics. In Phase I, we will investigate designs based on either coherent or incoherent illumination and downselect the most promising design for fabrication and optical characterization. We will also develop machine learning algorithms to quantify the reduction in computation time when inputting edge images versus raw (brightfield) images to a classification model. In Phase II, we will use inverse design and end-to-end techniques to co-optimize a meta-optic frontend and computational backend. We will then develop an advanced prototype for high-speed image classification and target recognition. Success in this effort will extend autonomous target recognition capabilities and enhance the lethality and survivability of the warfighter.