多任务衍射神经网络系统设计与实现  

Design and implementation of multi-task diffraction neural network system

作　　者：王子荣 张星祥[1] 龙勇机 付天骄[1] 张墨 WANG Zirong;ZHANG Xingxiang;LONG Yongji;FU Tianjiao;ZHANG Mo(Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033 [2]中国科学院大学,北京100049

出　　处：《液晶与显示》2024年第4期490-505,共16页Chinese Journal of Liquid Crystals and Displays

基　　金：国家自然科学基金青年科学基金(No.42001345);吉林省自然科学基金(No.20220101168JC)。

摘　　要：为探索利用衍射神经网络执行多任务图像分类识别的可行性,本文设计并搭建一种衍射神经网络系统。该系统采用空间光调制器(Spatial Light Modulator,SLM)做衍射神经网络的相位及振幅权重的调制以及网络层的光学全连接,并利用CMOS相机实现衍射神经网络中各衍射层输出的光电非线性激活与输出图像识别结果判别。设计的系统模型在MNIST和Fashion-MNIST图像分类识别中正确率达到94.1%和92.1%。最终通过搭建光路系统,光学实验正确率分别为91%和81.7%。所设计的衍射神经网络系统可满足多种图像分类识别应用需求,为衍射网络的设计与构建提供了新的思路。To investigate the feasibility of diffraction neural network to perform multi-task image classification recognition,a diffraction neural network system is designed and built.The system uses a spatial light modulator(SLM)to modulate the phase and amplitude weights of the diffraction neural network and the optical full connection of the network layers.A CMOS camera is adopted to realize the optical nonlinear activation of the output of each diffraction layer in the diffraction neural network and discriminate the output image recognition results.The designed system model achieves 94.1%and 92.1%accuracy in MNIST and Fashion-MNIST image classification recognition.Finally,by building optical path system,optical experiments have 91%and 81.7%accuracy respectively,which verifies that the designed diffraction neural network system can meet the requirements of various image.classification and recognition applications,and provides a new idea for the design and construction of diffraction networks.

关 键 词：衍射神经网络 光学神经网络 系统设计 图像分类识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]