计算机科学 ›› 2021, Vol. 48 ›› Issue (6A): 38-42.doi: 10.11896/jsjkx.201000160

• 图像处理&多媒体技术 • 上一篇    下一篇

结合MCycleGAN与RFCNN实现太阳斑点图高分辨重建

崔雯昊1, 蒋慕蓉1, 杨磊2, 傅鹏铭1, 朱凌霄1   

  1. 1 云南大学信息学院 昆明650500
    2 中国科学院云南天文台 昆明650011
  • 出版日期:2021-06-10 发布日期:2021-06-17
  • 通讯作者: 蒋慕蓉(jiangmr@ynu.edu.cn)
  • 作者简介:1207042764@qq.com
  • 基金资助:
    云南省高校科技创新团队支持项目(IRTSTYN);国家自然科学基金(11773073)

Combining MCycleGAN and RFCNN to Realize High Resolution Reconstruction of Solar Speckle Image

CUI Wen-hao1, JIANG Mu-rong1, YANG Lei2, FU Peng-ming1, ZHU Ling-xiao1   

  1. 1 School of Information Science and Technology,Yunnan University,Kunming 650500,China
    2 Yunnan Observatories,Chinese Academy of Sciences,Kunming 650011,China
  • Online:2021-06-10 Published:2021-06-17
  • About author:CUI Wen-hao,born in 1996,postgra-duate.His main research directions include deep learning and image reconstruction.
    JIANG Mu-rong,born in 1963,professor,Ph.D.Her main research directions include mathematical methods of image processing and intelligent calculation.
  • Supported by:
    Science and Technology Innovation Team Support Project of Yunnan Province (IRTSTYN) and National Natural Science Foundation of China (11773073).

摘要: 太阳斑点图高分辨率重建是天文图像处理的重要研究内容之一。基于深度学习的图像高分辨率重建,通过神经网络模型学习获得低分辨率图像到高分辨率图像的端到端映射函数,能够恢复图像高频信息,但在用于特征单一、噪音较多、局部细节模糊的太阳斑点图重建时,存在边缘过于平滑、高频信息易丢失等不足。将输入图像与重建图像结构特征加入CycleGAN网络中得到MCycleGAN,利用生成器网络从结构特征中获取高频信息,计算特征差来增强网络重建高频信息的能力;将残差块和融合层加入DeepFuse网络中构建RFCNN,利用图像帧间相似信息互补进行多帧重建,使重建图像边缘更加清晰。利用所提方法的重建结果与云南天文台使用的斑点掩膜法Level1+的结果对比表明,所提算法具有误差小、重建图像清晰度高等优点。

关键词: 太阳斑点图像, 深度学习, 高分辨率重建, MCycleGAN, RFCNN

Abstract: High resolution reconstruction of solar speckle image is one of the important research contents in astronomical image processing.High resolution image reconstruction based on deep learning can obtain the end-to-end mapping function from low-resolution image to high-resolution image through neural network model learning,which can recover the high-frequency information of the image.However,when reconstructing the sun speckle image with single feature,more noise and fuzzy local details,there are some shortcomings such as too smooth edge and easy loss of high-frequency information.In this paper,the structure features of input image and reconstructed image are added to CycleGAN network to get MCycleGAN.High frequency information is obtained from structural features by generator network,and the feature difference is calculated to enhance the ability of network to reconstruct high-frequency information.Residual block and fusion layer are added to DeepFuse network to construct RFCNN,and multi frame reconstruction is carried out by using similar information between image frames.The edge of the reconstructed image is clearer.The reconstruction result is compared with the speckle mask method Level1+ used by Yunnan Observatory,which shows that the proposed algorithm has the advantages of small error and high definition of reconstructed image.

Key words: Solar speckle image, Deep learning, High resolution reconstruction, MCycleGAN, RFCNN

中图分类号: 

  • TP391.41
[1] XIANG Y Y,LIU Z,JIN Z Y.High resolution solar image reconstruction method [J].Progress in astronomy,2016,1:94-110.
[2] HUO Z X,ZHOU J F.Method of astronomical image recon-struction from speckle pattern [J].Progress in Astronomy,2010(1):74-94.
[3] WU R G.Image super resolution reconstruction based on deep learning [D].Chengdu:University of Chinese Academy of Sciences (Institute of Optoelectronic Technology,Chinese Academy of Sciences),2020.
[4] GOODFELLOW I J,POUGET-ANADIE J.Genera-tive Adversarial Networks [J].Advances in Neural Information Processing Systems,2014(3):2672-2680.
[5] TANG C,LI J.Unpaired Low-Dose CT Denoising NetworkBased on Cycle-Consistent Generative Adversarial Network with Prior Image Information[J].Computational and Mathematical Methods in Medicine,2019(12):1-11.
[6] SIMONYAN K,ZISSERMAN A.Very Deep Convolutional Networks for Large-Scale Image Recognition[C]//Computer Vision and Pattern Recognition.2014.
[7] LEDIG C,THEIS L.Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network [C]//CVPR.2017:574-591.
[8] PRABHAKAR K R,SRIKAR V S,BABU R V.DeepFuse:A Deep Unsupervised Approach for Exposure Fusion with Ex-treme Exposure Image Pairs [C]//ICCV.2017:472.
[9] TAI Y,GUO Y,CHEN Q,et al.Auto-embedding generativeadversarialnetworks for high resolution image synthesis[J].IEEE Transactions on Multimedia,2019(6):112.
[10] ZENG R H ,XU H M,HUANG W B,et al.Dense regression networkfor video grounding[C]//IEEE Conference on Compu-ter Vision and Pattern Recognition.2020:618.
[11] LEDIG C,THEIS L,HUSZAR F,et al.Photo-Realistic SingleImage Super-Resolution Using a Generative Adversarial Network [C]//IEEE Conference on Computer Vision & Pattern Recognition(CVPR).2016:137.
[12] LEDIG C,THEIS L.Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network [C]//CVPR.2017:574-591.
[13] ZHANG S L,YI B S,LI W Z,et al.Multi focus image fusion method based on image matting technology [J].Computer Applications,2016,36(7):1949-1953.
[14] LIU Y,LIU S,WANG Z.Multi-focus image fusion with dense SIFT[J].Information Fusion,2015,23:139-155.
[15] BEVILACQUA M,ROUMY A,GUILLEMOT C,et al.Low-complexity single-image super-resolution based on nonnegative neighbor embedding[C]//The British Machine VisionConfe-rence(BMVC).2012.
[16] ZEYDE R,ELAD M,PROTTER M.On single image scale-up using sparse-representations[C]//Curves and Surfaces.Sprin-ger,2012:711-730.
[17] MARTIN D,FOWLKES C,TAL D,et al.A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics[C]//IEEE International Conference on Computer Vision (ICCV).2001:416-423.
[1] 冯芙蓉, 张兆功. 目标轮廓检测技术新进展[J]. 计算机科学, 2021, 48(6A): 1-9.
[2] 和青芳, 王慧, 程光. 自适应小数据集乳腺癌病理组织分类研究[J]. 计算机科学, 2021, 48(6A): 67-73.
[3] 刘汉卿, 康晓东, 李博, 张华丽, 冯继超, 韩俊玲. 利用深度学习网络对医学影像分类识别的比较研究[J]. 计算机科学, 2021, 48(6A): 89-94.
[4] 韩斌, 曾松伟. 基于多特征融合和卷积神经网络的植物叶片识别[J]. 计算机科学, 2021, 48(6A): 113-117.
[5] 刘荣, 张宁. 图片分析在电子商务中的应用现状与未来趋势——基于图片视觉和内容特征的研究综述[J]. 计算机科学, 2021, 48(6A): 137-142.
[6] 裴莹, 李天祥, 王鏖清, 付加胜, 韩霄松. 基于新闻的国际天然气价格趋势预测方法[J]. 计算机科学, 2021, 48(6A): 235-239.
[7] 牛康力, 谌雨章, 张龚平, 谭前程, 王绎冲, 罗美琪. 基于深度学习的无人机航拍车流量监测[J]. 计算机科学, 2021, 48(6A): 275-280.
[8] 杨进才, 曹元, 胡泉, 沈显君. 基于Transformer模型与关系词特征的汉语因果类复句关系自动识别[J]. 计算机科学, 2021, 48(6A): 295-298.
[9] 曾伟良, 陈漪皓, 姚若愚, 廖睿翔, 孙为军. 时空图注意力网络在交叉口车辆轨迹预测的应用[J]. 计算机科学, 2021, 48(6A): 334-341.
[10] 李达, 雷迎科, 张海川. 基于LTE网络的室外指纹定位[J]. 计算机科学, 2021, 48(6A): 404-409.
[11] 王丹妮, 陈伟, 羊洋, 宋爽. 基于高斯增强和迭代攻击的对抗训练防御方法[J]. 计算机科学, 2021, 48(6A): 509-513.
[12] 钱基德, 熊仁和, 王乾垒, 杜冬, 王在俊, 钱基业. 边缘计算在飞行训练中的应用[J]. 计算机科学, 2021, 48(6A): 603-607.
[13] 刘昱彤, 李鹏, 孙云云, 胡素君. 基于深度动态联合自适应网络的图像识别方法[J]. 计算机科学, 2021, 48(6): 131-137.
[14] 陈晋音, 邹健飞, 袁俊坤, 叶林辉. 面向恶意软件检测模型的黑盒对抗攻击方法[J]. 计算机科学, 2021, 48(5): 60-67.
[15] 张少钦, 杜圣东, 张晓博, 李天瑞. 融合多模态信息的社交网络谣言检测方法[J]. 计算机科学, 2021, 48(5): 117-123.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
[1] 方磊, 武泽慧, 魏强. 二进制代码相似性检测技术综述[J]. 计算机科学, 2021, 48(5): 1 -8 .
[2] 朱雨, 庞建民, 徐金龙, 陶小涵, 王军. 面向SW26010处理器的三维Stencil自适应分块参数算法[J]. 计算机科学, 2021, 48(6): 10 -18 .
[3] . 目录[J]. 计算机科学, 2021, 48(6A): 0 .
[4] 冯芙蓉, 张兆功. 目标轮廓检测技术新进展[J]. 计算机科学, 2021, 48(6A): 1 -9 .
[5] 孙正, 张小雪. 生物光声成像中声反射伪影抑制方法的研究进展[J]. 计算机科学, 2021, 48(6A): 10 -14 .
[6] 周欣, 刘硕迪, 潘薇, 陈媛媛. 自然交通场景中的车辆颜色识别[J]. 计算机科学, 2021, 48(6A): 15 -20 .
[7] 黄雪冰, 魏佳艺, 沈文宇, 凌力. 基于自适应加权重复值滤波和同态滤波的MR图像增强[J]. 计算机科学, 2021, 48(6A): 21 -27 .
[8] 江妍, 马瑜, 梁远哲, 王原, 李光昊, 马鼎. 基于分数阶麻雀搜索优化OTSU肺组织分割算法[J]. 计算机科学, 2021, 48(6A): 28 -32 .
[9] 张子丞, 谭志苇, 张晨瑞, 王旋, 刘晓璇, 俞一彪. 基于高低频带对数能量谱比贝叶斯决策的语音端点检测[J]. 计算机科学, 2021, 48(6A): 33 -37 .
[10] 田洋, 毕秀丽, 肖斌, 李伟生, 马建峰. 基于离散切比雪夫变换的图像接缝裁剪篡改检测[J]. 计算机科学, 2021, 48(6A): 43 -50 .