<p>Ghost imaging is an optical imaging method that obtains information by the correlation calculation of light intensity. However, ghost image technology in the encryption system needs to encrypt images one by one, which will cause slow speed and many inconveniences. Multi-image encryption technology can greatly improve the efficiency of image transmission but faces potential security risks. In this paper, a multi-image encryption method with light-field modulation, ghost imaging, and deep learning (L-G-D multi-image encryption) is proposed. This paper uses computational ghost imaging encryption and holographic encryption technology to increase encryption capacity. a series of normal Gaussian random distribution matrices is used as the primary key 1, Modulation matrices are selected at random from the primary key 1. Modulation matrices and Fresnel diffraction distance make up the Secondary key 2 together, which can reduce potential security risks. Finally, this paper develops a deep neural network based on the Pix2Pix framework to make the Super-Resolution of decrypted images and enhance the anti-noise ability. This paper improved the capacity of ghost imaging encryption and made information encryption more secure.</p>