Terahertz optical machine learning for object recognition

Abstract

We demonstrate an optical machine learning method in the terahertz domain, which allows the recognition of objects within a single measurement. As many materials are transparent in the terahertz spectral region, objects hidden within such materials can be identified. In contrast to typical object recognition methods, our method only requires a single pixel detector instead of a focal plane array. The core of the calculation is performed by a quantum cascade laser generated terahertz beam, which is spatially modulated at a near-infrared encoded silicon wafer. We show that this method is robust against displacements of the objects and noise. Additionally, the method is flexible and, due to the optically performed recognition task, inherently fast.