作為微波遙感的代表,SAR不僅具有覆蓋范圍大、包含信息量大、獲取信息快等一般遙感的特點(diǎn),而且具有全天時、全天候等不受光照、氣候環(huán)境影響的特點(diǎn),因此被廣泛應(yīng)用于國防安全建設(shè)和國民經(jīng)濟(jì)發(fā)展等眾多領(lǐng)域。其中,SAR變化檢測是SAR圖像解譯的關(guān)鍵組成部分,一直受到國內(nèi)外學(xué)者的廣泛關(guān)注。但是由于SAR圖像固有的成像機(jī)理,其不可避免的存在相干斑噪聲,對SAR變化檢測產(chǎn)生重要的影響。本文為了抑制相干斑噪聲對SAR變化檢測產(chǎn)生的影響,提高檢測精度,結(jié)合多尺度特征,利用深度自編碼等模型提取判別性特征,對SAR圖像變化檢測進(jìn)行研究。本文的研究內(nèi)容主要有以下四個方面:1.提出了一種快速無監(jiān)督深度融合的SAR圖像變化檢測框架（FuDFN）。其主要目的是利用棧式自動編碼器在特征學(xué)習(xí)過程中生成差異圖。與淺層網(wǎng)絡(luò)相比,該框架可以提取更多的有用特征,有利于獲得更好的變化檢測結(jié)果。此外,我們還找到了一個完整樣本的訓(xùn)練子集,它可以恰當(dāng)?shù)拇碚麄�數(shù)據(jù)集,既可以加速深度神經(jīng)網(wǎng)絡(luò)的訓(xùn)練,又可以避免欠擬合。而且,我們還設(shè)計了一個融合網(wǎng)絡(luò)結(jié)構(gòu),該結(jié)構(gòu)可以結(jié)合基于比值算子的方法,以確保較高層的表示優(yōu)于較低層的表示。對四幅真實(shí)合成孔徑... 

【文章來源】：西安電子科技大學(xué)陜西省 211工程院校 教育部直屬院校

【文章頁數(shù)】：155 頁

【學(xué)位級別】：博士

【文章目錄】：
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


【參考文獻(xiàn)】：
期刊論文
[1]神經(jīng)網(wǎng)絡(luò)七十年:回顧與展望[J]. 焦李成,楊淑媛,劉芳,王士剛,馮志璽.  計算機(jī)學(xué)報. 2016(08)
[2]稀疏認(rèn)知學(xué)習(xí)、計算與識別的研究進(jìn)展[J]. 焦李成,趙進(jìn),楊淑媛,劉芳,謝雯.  計算機(jī)學(xué)報. 2016(04)
[3]基于復(fù)Bandelets的自適應(yīng)SAR圖像相干斑抑制[J]. 楊曉慧,焦李成,李登峰.  電子學(xué)報. 2009(09)

博士論文
[1]基于Fisher分類器和計算智能的遙感圖像變化檢測[D]. 辛芳芳.西安電子科技大學(xué) 2011



本文編號：3302121