Optical encryption-transmission via computational ghost imaging and fractional OAM multiplexing

This paper proposes a solution that combines computational ghost imaging (CGI), fractional orbital angular momentum (FOAM), and deep learning for the secure encryption and transmission of optical images. In the information collection and encryption stage relying on CGI, the bucket signals of the plaintext image are reindexed, while those of the reference image are converted into a sparse sequence, thus forming an encrypted signal set. During the FOAM encoding and transmission stage, these signals are mapped onto superposed vortex beam modes (256 modes for 8-bit data). A rotational phase shift is added to strengthen the system's security. For decryption, a pre-trained DenseNet multi-label classification network is utilized. This network can accurately identify the superposed topological charges of FOAM from the light intensity patterns captured by the CCD at the receiving end. To further elevate the security level, an authentication mechanism has been introduced. This "authentication-decryption" approach can prevent unauthorized access and ensure the legality of signals. This solution not only integrates the strengths of CGI in encryption with the high-dimensional transmission characteristics of FOAM but also presents a practical avenue for interdisciplinary exploration at the crossroads of OAM-based optical communication and CGI-based information security.