Diffractive deep neural network-based chaotic encryption system

In this Letter, we propose an optical image encryption system based on the Diffractive Deep Neural Network (DDNN) integrating the Logistic Chaotic Map (LCM) and two-dimensional Linear Canonical Transform (2D-LCT), to overcome static phase design and discrete modules in conventional optical encryption. The LCM performs pixel permutation, and 2D-LCT realizes transform-domain global diffusion. By optimizing diffractive layer phases, the hybrid encryption mapping is integrated into optical diffraction and modulation. The system is highly sensitive to key parameters. Simulations verified favorable reconstruction and generalization, proving the value and feasibility of DDNN-based optical encryption.