Optical classification of images at different wavelengths using spectral diffractive neural networks

Abstract

A solution of several different problems of image classification at several different wavelengths using a diffractive neural network (DNN) consisting of sequentially located phase diffractive optical elements (DOEs) is considered. To solve the classification problems, the problem of calculating the DNN is formulated as that of minimizing a functional that depends on the functions of the DOE diffractive microrelief heights - which form a DNN - and represents an error in solving the classification problems in question at the operating wavelengths. Explicit expressions are obtained for the functional derivatives and on this basis, a gradient method for calculating the DNN is formulated. Using the proposed gradient method, DNNs are calculated intended for solving three different problems of image classification at three different wavelengths. The presented simulation results of the calculated DNNs demonstrate their good performance characteristics and confirm the good performance of the proposed method.

Keywords

• image classification problem
• diffractive neural network
• cascaded diffractive optical element
• diffractive microrelief
• scalar diffraction theory
• optimization
• gradient method