尚光网 本站首页 本所首页 联系我们
chi20181129上海光机所司徒国海557456Computational optical imaging technology is a combination of traditional optical imaging and computer science, which further enriches the means of optical imaging. The technology has made a lot of progress in improving imaging resolution, phase recovery, 3D imaging, holography, ghost imaging and imaging through scatter media. As an emerging technology in recent years, deep learning has been widely applied to many fields including chess games, biometrics, and self-driving car. However, in the field of optical imaging, deep learning is just beginning to emerge. This thesis explores several applications of deep learning technology in the field of optical imaging, mainly in the following three aspects: Firstly, exploring the application of deep learning in computational ghost imaging. The proposed method utilizes deep learning to denoise a low signal sampling ratio (SNR) images recovered by the second-order intensity correlation algorithm, and obtains a restored image with high SNR. This method can recover good results with extremely low SNR, greatly reducing the number of signal samples during computational ghost imaging experiments. Besides,the proposed method can work under a lower signal-to-sampling ratio than compressive ghost imaging. Sencondly, exploring the application of deep learning in imaging through scattering media. The proposed method utilizes deep learning to retrieve the image of the object hidden behind scattering media. In addition, it does not require to measuring the complex amplitude, but only the input-output intensities, which simplifies the experimental system setup. As a new method, deep learning opens up a new path to explore optical imaging through more general complex systems. Thirdly, exploring the application of deep learning in digital holography. The proposed method utilizes deep learning to reconstruct the object wavefront from a single-shot in-line hologram, with the removal of the unwanted zero-order and twin image terms. The reconstruction is a one-step process without the need of back-propagation. Then, we also proposed a fast autofocusing method in digital holography using the magnitude differential, and found that digital holography can be applied to the mode decomposition for optical fibers. In this dissertation, we mainly discuss the preliminary application of deep learning in ghost imaging, imaging through scattering media and digital holography. The feasibility of deep learning in these fields is verified by numerical simulation and corresponding experiments. It lays the foundation for deep learning in more complex and practical applications of optical imaging.learning in these fields is verified by numerical simulation and corresponding experiments. It lays the foundation for deep learning in more complex and practical applications of optical imaging.2019atalunwen2191211145716388Computing Optical Imaging;Deep Learning;Ghost Imaging;Imaging through Scattering Media;Digital HolographyOn the Applications of Deep Learning in Computing Optical Imaging深度学习在计算光学成像中的若干应用传统光学成像与计算机技术相结合,产生了计算光学成像技术,它进一步丰富了光学成像的手段,并且在提高成像分辨率,相位恢复,三维成像,全息,鬼成像和散射成像等方面都取得了很多进展。 深度学习作为近几年的一种新兴技术,已经被广泛应用到包括棋类博弈,生物特征识别,无人驾驶在内的诸多领域,而在光学成像领域里,深度学习才刚刚开始崭露头角。本论文探究了深度学习技术在光学成像领域的若干应用,主要包括以下三个方面: 一,探索深度学习在计算鬼成像中的应用。我们提出的方法利用深度学习给二阶强度关联算法恢复出的鬼成像的低信噪比图像去噪,获取高信噪比的恢复图像。该方法能够在极低信号采样比的情况下恢复出较好的结果,极大的减少了计算鬼成像实验过程中的信号采样次数。同常用的压缩感知鬼成像技术相比,本方法可以在更低的信号采样比条件下工作。 二,探索深度学习在散射成像中的应用。我们提出的方法利用深度学习从实验采集到的散斑图像中恢复出被散射介质遮挡的目标物体,该方法只需要单张散斑图便可以透过较厚的散射样本恢复出目标物体。另外,该方法不需要测量复振幅,只需要得到输入输出的强度,不需要干涉装置,简化了实验装置。作为一种新方法,深度学习给我们开辟了一条研究光学散射成像问题的新道路。 三,探索深度学习在数字全息术中的应用。我们提出的方法利用深度学习技术直接去除数字全息图里的直流项和孪生像,从单幅全息图获得目标物体的光场,其重构过程一步完成,无需反向传播。另外,在研究数字全息的同时,我们还提出了一种基于轴向差分的数字全息对焦技术,发现数字全息可应用于光纤的模式分解。 本论文主要初步讨论了深度学习在鬼成像、散射成像和数字全息方面的相关应用。从数值仿真到对应的实验,我们验证了深度学习在这些领域应用的可行性,这为深度学习今后在光学成像领域里更复杂更实用的应用打下了基础。计算光学成像;深度学习;鬼成像;散射成像;数字全息中国科学院上海光学精密机械研究所吕萌光学工程博士
中文题目: 深度学习在计算光学成像中的若干应用
外文题目: On the Applications of Deep Learning in Computing Optical Imaging
作者: 吕萌
导师姓名: 司徒国海
学位授予机构: 中国科学院上海光学精密机械研究所
答辩时间: 20181129
中文关键词:
计算光学成像;深度学习;鬼成像;散射成像;数字全息
英文关键词:
Computing Optical Imaging;Deep Learning;Ghost Imaging;Imaging through Scattering Media;Digital Holography
中文摘要:
英文摘要:
文献类型:学位论文
学位级别: 博士
正文语种: chi
全文传递服务
clickdetails
页面点击量: 35
文章下载量: 1
visitlog
友情链接:
  上海大恒公司
  南京先进激光技术院
  光学产品库
版权所有 © 2009 中国科学院上海光学精密机械研究所 沪ICP备05015387号
主办:中国科学院上海光学精密机械研究所 上海市嘉定区清河路390号(201800)