Terahertz compressive optical-digital neural network imaging

The Terahertz (THz) frequency band has garnered significant attention due to its superior performance in non-invasive sensing and imaging. However, current THz imaging systems face challenges related to hardware limitations, resulting in information loss during the data digitization process and insufficient image throughput. To address these issues, we propose a hybrid compressed optical-digital neural network designed to facilitate real-time THz imaging and precise extraction of object information. This method employs an Optical Neural Network (ONN) as a physical encoder to reduce and compress the signal, adapting to the physical constraints of THz sensors. Through the joint training of a Digital Neural Network (DNN), the compressed signals can be reconstructed, showcasing high-quality images, an expanded field of view, and real-time video capture capabilities without requiring additional motor systems for raster scanning. The Terahertz (THz) frequency band has recently received widespread attention for its outstanding capabilities in non-invasive and non-destructive sensing and imaging applications. However, current THz imaging systems face significant challenges due to hardware limitations, resulting in information loss during the data digitization and information extraction processes, as well as restricted image throughput. To overcome these challenges, we propose a hybrid compressed optical-digital neural network aimed at facilitating real-time THz imaging and accurate extraction of object information. This method utilizes a physical encoder, namely the Optical Neural Network (ONN), to convert and reduce the dimensionality of the physical signal, effectively compressing it to meet the physical limits of THz sensor arrays. Once the compressed signal is captured and digitized by the THz sensor array, the jointly trained Digital Neural Network (DNN) reconstructs the signal back to its desired or original form. Our proposed THz ONN-DNN computational imaging system demonstrates enhanced image quality, an extended field of view (without a lens system), diffraction-free imaging capabilities, and real-time THz video capture at two frames per second. © 2025 All rights reserved.