Diffractive deep neural network motivating high-performance optical information encryption

In this Letter, a diffractive deep neural network (DDNN) optical system has been proposed to implement a discrete fractional Fourier transform (DFrFT). By optimizing the phase distributions of the successive diffractive layers, the designed DDNN optical system can accurately implement DFrFT either with single order or with multiple orders simultaneously at different output planes. Further, cooperating with the Arnold scrambling algorithm, the proposed DDNN-based DFrFT system shows very sensitive characteristics for the offset of the receiving position of the output plane, the mismatch of the fractional order between the DFrFT and inversion DFrFT, and the parameters of the Arnold scrambling algorithm simultaneously. This security characteristic provides a feasible solution for optical information encryption.