CASIA OpenIR  > 毕业生  > 硕士学位论文
基于深度学习的电镜图像超分辨率技术研究与应用
杨茂柯1,2
Subtype工程硕士
Thesis Advisor韩华
2018-06
Degree Grantor中国科学院研究生院
Place of Conferral北京
Keyword扫描电镜 深度学习 图像超分辨率重建
Abstract
本研究希望能够通过牺牲电镜图像的拍摄质量来加速电镜图像的采集过程,在采集完成后再使用超分辨率重建算法以后处理的方式恢复图像质量。因为拍摄低分辨率图像消耗的时间远小于拍摄高分辨率图像消耗的时间,从而可以成倍数地加速数据的采集。
 
因为加速拍摄引入的噪声会对常规训练的超分辨率重建模型产生严重的干扰。针对于这一干扰,本研究提出先对噪声进行合理建模,再根据建模结果生成恰当的训练数据,用于超分辨率重建模型的训练。这一方案的提出使得现有的绝大多数基于深度学习的超分辨率重建模型都可以用于本项应用。针对于该方案,本研究提出了两种对噪声建模的方式,均能够对电镜图像的噪声进行准确的建模。最后本研究提出了一种针对于电镜图像的超分辨率重建模型,并使用另外两种经典的超分辨率重建模型按照被研究提出的技术路线实现了对于带噪电镜图像的超分辨率重建。实验结果均显示了本研究中提出的方法的有效性。
Other Abstract
We propose to employ image super resolution to  accelerate collection speed of scanning electric microscopes(SEM). This process can be done by collecting images in a relatively lower resolution, and then upscale the collected images with image super-resolution(SR) algorithms. For the SEM-images, current state-of-the-art methods are good to use. However, because of the purpose of employ SR algorithms to deal with SEM images is to accelerate the collection process, some difference are included in the application.
 
There are ome huge difference between the normal nature images and SEM images. First of all, for a same size of an image patch, SEM images have less useful information than nature images. Thus, structure and details of SEM images may need larger receptive field to represent. Besides, the noise in SEM images cannot be ignored, otherwise it will disturb the reconstruction process of SR algorithm.
 
The characteristic of SEM images makes it hard to obtain suitable training dataset to train a SR model mapping between noised low resolution images and unnoised high resolution images. In order to bridge the gap, two kind of methods is proposed, and named Noise-GAN and NENet respectively. And a pipeline is proposed to generate suitable training data for SR models. Finally,  two state-of-the-art SR models is employed, and one new SR model is designed to do experiments on our application. Results illustrate that our pipeline works well on SEM-image datasets.
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/21512
Collection毕业生_硕士学位论文
Affiliation1.中国科学院大学
2.中国科学院自动化研究所
First Author AffilicationInstitute of Automation, Chinese Academy of Sciences
Recommended Citation
GB/T 7714
杨茂柯. 基于深度学习的电镜图像超分辨率技术研究与应用[D]. 北京. 中国科学院研究生院,2018.
Files in This Item:
File Name/Size DocType Version Access License
杨茂柯-毕业论文.pdf(14435KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[杨茂柯]'s Articles
Baidu academic
Similar articles in Baidu academic
[杨茂柯]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[杨茂柯]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.