摘要
Terahertz (THz) waves are utilized in various imaging systems due to their unique optical properties, enabling penetration of non-metallic and non-polar materials without ionization. However, limitations of sensors and severe diffractive effects constrain the application of THz imaging. To address these challenges, we propose THz computational imaging, employing an all-optical neural network (AONN) and an electric neural network (ENN) to construct a variational autoencoder (VAE). The AONN compresses the signal below the sensor's limitations, while the ENN extracts high-quality imaging from the compressed data. To obtain imaging that surpasses sensor limitations and mitigates diffractive effects in THz imaging.