近年来,深度学习技术的爆发式发展引领了机器学习的又一次浪潮。深度神经网络具备抽象特征的高效识别与提取能力、强大的非线性拟合能力、抗干扰鲁棒性及非凡的泛化能力,被广泛应用于自动驾驶、目标识别、机器翻译、语音识别等领域。最近几年,卷积神经网络(CNN)在光学信息处理中获得广泛应用,本文介绍CNN的基础概念和结构构成,回顾其在数字全息术、条纹分析、相位解包裹、鬼成像、傅里叶叠层成像、超分辨显微成像、散射介质成像、光学层析成像等领域的最新应用进展,评述CNN在光学信息处理中的典型应用特点,最后分析CNN应用于光学信息处理中的不足,并展望其未来发展。

The explosive development of deep learning technology has led another wave of machine learning in recent years. Deep neural network, with the ability to recognize and extract abstract features, fit nonlinear relationships, against interference factors and generalization, is widely used in autopilot, target recognition, machine translation, speech recognition and other fields. The convolutional neural networks (CNN) are popular in optical information processing. In this paper, we introduce the basic concepts and structural components of CNN in detail, and review the applications in digital holography, fringe patterns analysis, phase unwrapping, ghost imaging, Fourier ptychographic microscopy, super-resolution microscopy, scattering medium imaging, optical tomography imaging, etc. We summarize the typical applications and existing shortages of CNN in optical information processing, and finally prospect the future development of convolutional neural networks.