Multi-wavelength diffractive neural network with the weighting method

This is the authors' abstract. We don't add key points for this paper.

Recently, the diffractive deep neural network (D 2 NN) has demonstrated the advantages to achieve large-scale computational tasks in terms of high speed, low power consumption, parallelism, and scalability. A typical D 2 NN with cascaded diffractive elements is designed for monochromatic illumination. Here, we propose a framework to achieve the multi-wavelength D 2 NN (MW-D 2 NN) based on the method of weight coefficients. In training, each wavelength is assigned a specific weighting and their output planes construct the wavelength weighting loss function. The trained MW-D 2 NN can implement the classification of images of handwritten digits at multi-wavelength incident beams. The designed 3-layers MW-D 2 NN achieves a simulation classification accuracy of 83.3%. We designed a 1-layer MW-D 2 NN. The simulation and experiment classification accuracy are 71.4% and 67.5% at RGB wavelengths. Furthermore, the proposed MW-D 2 NN can be extended to intelligent machine vision systems for multi-wavelength and incoherent illumination.