KMS of Xinjiang Astronomical Observatory, CAS
| WARSHIP：用于天文图像超分辨率的卷积神经网络算法 | |
| Alternative Title | WARSHIP: The convolutional neural network algorithms for Astronomical Image Super-Resolution |
许文帝
| |
| Subtype | 硕士 |
| Thesis Advisor | 张明 |
| 2018 | |
| Degree Grantor | 中国科学院大学 |
| Place of Conferral | 北京 |
| Degree Name | 理学硕士 |
| Degree Discipline | 天体物理 |
| Keyword | 深度学习,天文成像,类脑计算 |
| Abstract | 不论是光学还是射电望远镜，都需要天文图像超分辨率（AISR）来打破物理仪器的硬件分辨率瓶颈，从而获得更高的角分辨率，进而取得天文学科乃至于物理学科上的科学发现。最近比较成功的用于AISR的算法，包括传统的统计方法l1+TV、基于机器学习的算法CHIRP和我们的WARSHIP-XZNets。在机器学习的子集深度学习的框架下，我们首次总结了用于机器视觉的5个重要理念，并用WARSHIP来表示，同时强调了它们源于脑科学的灵感。在WARSHIP，特别是其中的脑科学理念的指导下，我们设置了一系列模型/算法，叫做WARSHIP-XZNets，用于AISR，它们都被证明是比较成功的类脑计算的实现。我们的WARSHIP-XZNets在处理通用自然图片集 Set5（在深度学习社团中广泛使用的基准）时，从精度和速度两方面获得了一个很好的平衡，展示了在通用自然图片集上的优越适用性。当我们把特定的天文图片送给我们的算法去处理时，算法的表现也很好。在重建超大质量黑洞的吸积盘图片时，PSNR数值相当高，高达44.35！频率论探测上限在5-sigma时，对M31图像的致密特征的探测的置信区间是80\%。从而，我们证明了我们的WARSHIP-XZNets在AISR中的强鲁棒性。 |
| Other Abstract | Both optical and radio telescopes need astronomical image super-resolution(AISR) to break the resolution limitation imposed by physical instrument to pursue higher angular resolution for scientific discoveries in astronomy and physics. Recently successful algorithms for AISR include traditional statistical algorithm, l1+TV , and machine learning based algorithms, like CHIRP and our WARSHIP-XZNets.Within a subset of machine learning --deep learning framework, to our best knowledge, we firstly summarize 5 key ideas for machine vision, which are represented as WARSHIP, and emphasize their inspirations from brainscience. Following WARSHIP, especially the biological ideas, we deploy algorithms of WARSHIP-XZNets for AISR ,which prove to be successful implementations of brain-inspired computing (BIC).WARSHIP-XZNets perform a happy medium between performance and speed on Set5, a benchmark wildely used in deep learning community, showing its nice applicability for general natural images. When we send specified astronomical images to our pipeline , it also works well. The Peak Signal Noise Retio (PSNR) of recontructed accretion disk image of super massive black hole is pretty high, 44.35! The confidence level of the frequentist upper limit for a 5-sigma detection of the compact feature in M31 image is 80\%. Therefore, we demonstrate strong robustness of our BIC-- WARSHIP-XZNets for AISR. |
| Pages | 54 |
| Language | 中文 |
| Document Type | 学位论文 |
| Identifier | http://ir.xao.ac.cn/handle/45760611-7/4147 |
| Collection | 研究生学位论文 |
| Affiliation | 中国科学院新疆天文台 |
| First Author Affilication | Xinjiang Astronomical Observatory, Chinese Academy of Sciences |
| Recommended Citation GB/T 7714 | 许文帝. WARSHIP：用于天文图像超分辨率的卷积神经网络算法[D]. 北京. 中国科学院大学,2018. |
| Files in This Item: | ||||||
| File Name/Size | DocType | Version | Access | License | ||
| 许文帝.pdf（13851KB） | 学位论文 | 暂不开放 | CC BY-NC-SA | Application Full Text | ||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment