Machine learning assisted beyond the diffraction limit imaging

机器学习辅助超越衍射极限成像

• 批准号: 548995-2019
• 负责人: Gholipour, Behrad
• 金额: $ 1.46万
• 依托单位: University of Alberta
• 依托单位国家: 加拿大
• 项目类别: Alliance Grants
• 财政年份: 2019
• 资助国家: 加拿大
• 起止时间: 2019-01-01 至 2020-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=677204
• 关键词: Machine; learning; assisted; beyond; diffraction

We propose that a deep neural network and employment of various machine learning algorithms can significantly improve optical microscopy, enhancing its spatial resolution over a large field of view and depth of field. After its training, the only input to such a network can be an image acquired using a regular optical microscope, without any changes to its design. We will train our system using datasets generated by imaging a carefully designed and nanofabricated set of reference samples containing different materials and geometries, imaged above and below the optical diffraction limit using a combination of optical and electron microscopy. Post training, we will blindly test this approach using various samples that are imaged with low-resolution and wide-field systems. Our approach will allow us to rapidly output an image which surpasses the resolution capability of the system. This will allow us to maintain resolution while significantly surpassing the limited field of view and depth of field of conventional optical microscopy systems. Therefore, this project will culminate in the development of a universal machine learning adapter that will allow the conversion of any ordinary optical microscope into a wide field-of-view beyond the diffraction limit optical nanoscope.The proposed research program will have huge significance to any field that use microscopy tools, including, e.g., life and material sciences, where optical microscopy is considered as one of the most widely used and deployed techniques. Beyond that, the presented approach can be enhanced by using multispectral imaging, spanning different parts of the electromagnetic spectrum, and can be used to design computational imagers that get better as they continue to image different samples.

我们提出，深度神经网络和各种机器学习算法的使用可以显着改善光学显微镜，提高其在大视场和景深上的空间分辨率。经过训练后，此类网络的唯一输入可以是使用常规光学显微镜获取的图像，而无需对其设计进行任何更改。我们将使用通过对一组精心设计和纳米制造的包含不同材料和几何形状的参考样本进行成像而生成的数据集来训练我们的系统，使用光学和电子显微镜的组合在光学衍射极限之上和之下成像。训练后，我们将使用低分辨率和宽视场系统成像的各种样本盲目地测试这种方法。我们的方法将使我们能够快速输出超出系统分辨率能力的图像。这将使我们能够保持分辨率，同时显着超越传统光学显微镜系统的有限视场和景深。因此，该项目最终将开发出一种通用的机器学习适配器，该适配器将允许将任何普通光学显微镜转换为超出衍射极限的宽视场光学纳米显微镜。所提出的研究计划将对任何人都具有巨大的意义。使用显微镜工具的领域，包括生命和材料科学，其中光学显微镜被认为是最广泛使用和部署的技术之一。除此之外，所提出的方法可以通过使用多光谱成像来增强，跨越电磁频谱的不同部分，并且可以用于设计计算成像器，当它们继续对不同样本进行成像时，该方法会变得更好。