基于深度自编码框架的合成孔径雷达图像变化检测
发布时间：2021-07-25 13:33
作为微波遥感的代表,SAR不仅具有覆盖范围大、包含信息量大、获取信息快等一般遥感的特点,而且具有全天时、全天候等不受光照、气候环境影响的特点,因此被广泛应用于国防安全建设和国民经济发展等众多领域。其中,SAR变化检测是SAR图像解译的关键组成部分,一直受到国内外学者的广泛关注。但是由于SAR图像固有的成像机理,其不可避免的存在相干斑噪声,对SAR变化检测产生重要的影响。本文为了抑制相干斑噪声对SAR变化检测产生的影响,提高检测精度,结合多尺度特征,利用深度自编码等模型提取判别性特征,对SAR图像变化检测进行研究。本文的研究内容主要有以下四个方面:1.提出了一种快速无监督深度融合的SAR图像变化检测框架（FuDFN）。其主要目的是利用栈式自动编码器在特征学习过程中生成差异图。与浅层网络相比,该框架可以提取更多的有用特征,有利于获得更好的变化检测结果。此外,我们还找到了一个完整样本的训练子集,它可以恰当的代表整个数据集,既可以加速深度神经网络的训练,又可以避免欠拟合。而且,我们还设计了一个融合网络结构,该结构可以结合基于比值算子的方法,以确保较高层的表示优于较低层的表示。对四幅真实合成孔径...
【文章来源】：西安电子科技大学陕西省 211工程院校 教育部直属院校
【文章页数】：155 页
【学位级别】：博士
【文章目录】：
ABSTRACT
摘要
SYMBOL LIST
ABBREVIATION LIST
Chapter 1 Introduction
1.1 Research Significance and Background of the Subject
1.1.1 Remote Sensing and Remote Sensing Image
1.1.2 Development of SAR
1.1.3 Characteristics and Application of SAR Image
1.1.4 SAR Change Detection
1.2 Research Status and Difficulties of SAR Change Detection
1.2.1 Research Status of SAR Change Detection
1.2.2 Difficulties of SAR Change Detection
1.3 Deep Auto-encoder Framework
1.4 The Main Work and Content of This Paper
Chapter 2 Fast Unsupervised Deep Fusion Network for Change Detection of Multi-temporal SAR Images
2.1 Introduction
2.2 Fast Unsupervised Deep Neural Network for Change Detection
2.2.1 Establishment of Stacked Auto-encoder Network
2.2.2 Speeding up the Training of Deep Neural Network
2.3 Deep Fusion Network
2.4 Experiments and Analysis
2.4.1 Introduction to Data Sets and Evaluation Criteria
2.4.2 Analysis of Speed about FuDFN
2.4.3 Performance of FuDFN
2.4.4 Analysis of Parameters
2.4.5 Robust Analysis
2.5 Conclusion
Chapter 3 Feature Learning and Change Feature Classification based on VariationalAuto-encoder for SAR Change Detection
3.1 Introduction
3.2 Foundation of Related Network
3.2.1 Introduction to AE
3.2.2 Introduction to VAE
3.3 Change Detection based on SVAE
3.3.1 Preprocessing and FCM
3.3.2 SVAE for Feature Learning and Classification
3.4 Experimental Settings and Results Analysis
3.4.1 Data Description
3.4.2 General Information
3.4.3 Parameters Analysis
3.4.4 Analysis of Representation Ability
3.4.5 Comparison with Other Methods
3.5 Conclusion
Chapter 4 Learning Spatial-Temporal Features via a Recurrent Convolutional SiameseAuto-encoder for SAR Image Change Detection
4.1 Introduction
4.2 Change Detection based on Recurrent Convolutional Siamese Auto-encoderNetwork
4.2.1 Spatial Feature Extraction via the Convolutional Siamese Auto-encoder
4.2.2 Enhancing the Discrimination Features by Modeling Temporal De-pendency via Recurrent Sub-network
4.2.3 Fine-tuning of the Whole Network
4.3 Experimental Settings
4.3.1 Data sets
4.3.2 Evaluation Criteria and Compared Algorithms
4.3.3 Parameters Setting
4.4 Experimental Results and Analysis
4.4.1 Analysis of parameters
4.4.2 Results on Ottawa Data Set
4.4.3 Results on Red River Data Set
4.4.4 Results on Large Size Data Set
4.4.5 Robustness Results on Simulated Images
4.5 Conclusion
Chapter 5 Multiscale Visual Cognitive Network for SAR Image Change Detection
5.1 Introduction
5.2 Motivation
5.2.1 ReCNN
5.2.2 Modeling Human Visual Cognitive Scenario
5.3 Multiscale Visual Cognitive Network
5.3.1 Multiscale Spatial Feature Learning via Visual Block
5.3.2 Modeling Multiscale Temporal Dependency via Cognitive Block
5.4 Experiments
5.4.1 Data Description
5.4.2 General Information
5.4.3 Parameters Analysis
5.4.4 Analysis of Multiscale Spatial Temporal Feature
5.4.5 Comparison with Other Methods
5.5 Conclusion
Chapter 6 Conclusion and Future Work
6.1 Conclusion
6.2 Future Work
BIBLIOGRAPHY
ACKNOWLEDGEMENTS
RESUME
【参考文献】：
期刊论文
[1]神经网络七十年:回顾与展望[J]. 焦李成,杨淑媛,刘芳,王士刚,冯志玺. 计算机学报. 2016(08)
[2]稀疏认知学习、计算与识别的研究进展[J]. 焦李成,赵进,杨淑媛,刘芳,谢雯. 计算机学报. 2016(04)
[3]基于复Bandelets的自适应SAR图像相干斑抑制[J]. 杨晓慧,焦李成,李登峰. 电子学报. 2009(09)
博士论文
[1]基于Fisher分类器和计算智能的遥感图像变化检测[D]. 辛芳芳.西安电子科技大学 2011
本文编号：3302121
【文章来源】：西安电子科技大学陕西省 211工程院校 教育部直属院校
【文章页数】：155 页
【学位级别】：博士
【文章目录】：
ABSTRACT
摘要
SYMBOL LIST
ABBREVIATION LIST
Chapter 1 Introduction
1.1 Research Significance and Background of the Subject
1.1.1 Remote Sensing and Remote Sensing Image
1.1.2 Development of SAR
1.1.3 Characteristics and Application of SAR Image
1.1.4 SAR Change Detection
1.2 Research Status and Difficulties of SAR Change Detection
1.2.1 Research Status of SAR Change Detection
1.2.2 Difficulties of SAR Change Detection
1.3 Deep Auto-encoder Framework
1.4 The Main Work and Content of This Paper
Chapter 2 Fast Unsupervised Deep Fusion Network for Change Detection of Multi-temporal SAR Images
2.1 Introduction
2.2 Fast Unsupervised Deep Neural Network for Change Detection
2.2.1 Establishment of Stacked Auto-encoder Network
2.2.2 Speeding up the Training of Deep Neural Network
2.3 Deep Fusion Network
2.4 Experiments and Analysis
2.4.1 Introduction to Data Sets and Evaluation Criteria
2.4.2 Analysis of Speed about FuDFN
2.4.3 Performance of FuDFN
2.4.4 Analysis of Parameters
2.4.5 Robust Analysis
2.5 Conclusion
Chapter 3 Feature Learning and Change Feature Classification based on VariationalAuto-encoder for SAR Change Detection
3.1 Introduction
3.2 Foundation of Related Network
3.2.1 Introduction to AE
3.2.2 Introduction to VAE
3.3 Change Detection based on SVAE
3.3.1 Preprocessing and FCM
3.3.2 SVAE for Feature Learning and Classification
3.4 Experimental Settings and Results Analysis
3.4.1 Data Description
3.4.2 General Information
3.4.3 Parameters Analysis
3.4.4 Analysis of Representation Ability
3.4.5 Comparison with Other Methods
3.5 Conclusion
Chapter 4 Learning Spatial-Temporal Features via a Recurrent Convolutional SiameseAuto-encoder for SAR Image Change Detection
4.1 Introduction
4.2 Change Detection based on Recurrent Convolutional Siamese Auto-encoderNetwork
4.2.1 Spatial Feature Extraction via the Convolutional Siamese Auto-encoder
4.2.2 Enhancing the Discrimination Features by Modeling Temporal De-pendency via Recurrent Sub-network
4.2.3 Fine-tuning of the Whole Network
4.3 Experimental Settings
4.3.1 Data sets
4.3.2 Evaluation Criteria and Compared Algorithms
4.3.3 Parameters Setting
4.4 Experimental Results and Analysis
4.4.1 Analysis of parameters
4.4.2 Results on Ottawa Data Set
4.4.3 Results on Red River Data Set
4.4.4 Results on Large Size Data Set
4.4.5 Robustness Results on Simulated Images
4.5 Conclusion
Chapter 5 Multiscale Visual Cognitive Network for SAR Image Change Detection
5.1 Introduction
5.2 Motivation
5.2.1 ReCNN
5.2.2 Modeling Human Visual Cognitive Scenario
5.3 Multiscale Visual Cognitive Network
5.3.1 Multiscale Spatial Feature Learning via Visual Block
5.3.2 Modeling Multiscale Temporal Dependency via Cognitive Block
5.4 Experiments
5.4.1 Data Description
5.4.2 General Information
5.4.3 Parameters Analysis
5.4.4 Analysis of Multiscale Spatial Temporal Feature
5.4.5 Comparison with Other Methods
5.5 Conclusion
Chapter 6 Conclusion and Future Work
6.1 Conclusion
6.2 Future Work
BIBLIOGRAPHY
ACKNOWLEDGEMENTS
RESUME
【参考文献】：
期刊论文
[1]神经网络七十年:回顾与展望[J]. 焦李成,杨淑媛,刘芳,王士刚,冯志玺. 计算机学报. 2016(08)
[2]稀疏认知学习、计算与识别的研究进展[J]. 焦李成,赵进,杨淑媛,刘芳,谢雯. 计算机学报. 2016(04)
[3]基于复Bandelets的自适应SAR图像相干斑抑制[J]. 杨晓慧,焦李成,李登峰. 电子学报. 2009(09)
博士论文
[1]基于Fisher分类器和计算智能的遥感图像变化检测[D]. 辛芳芳.西安电子科技大学 2011
本文编号：3302121
本文链接：https://www.wllwen.com/shoufeilunwen/xxkjbs/3302121.html
