深度學(xué)習(xí)（DL）神經(jīng)網(wǎng)絡(luò)方法成為遙感領(lǐng)域研究的熱點。深度學(xué)習(xí)神經(jīng)網(wǎng)絡(luò)是近年來的一項發(fā)展,已成為計算機(jī)視覺和遙感學(xué)科研究的課題。超分辨率重建是指通過數(shù)字圖像處理從單個或一系列低分辨率圖像中重建高分辨率圖像的技術(shù)。該技術(shù)不僅可以增加圖像的高頻信息,還可以消除低分辨率。不同的衛(wèi)星數(shù)據(jù)被用來預(yù)測每個深度學(xué)習(xí)模型的性能。深度學(xué)習(xí)在現(xiàn)代數(shù)字圖像處理方面取得了突破性進(jìn)展。與傳統(tǒng)算法的Bicubic和最大似然（ML）相比,圖像分類和目標(biāo)檢測等一系列具有挑戰(zhàn)性的圖像處理問題需要找到快速可靠的解決方案,因此,我們的論文的焦點是在遙感領(lǐng)域的各個重要階段應(yīng)用深度學(xué)習(xí)方法后,主要如下:1.超分辨率重建是指通過數(shù)字圖像處理從單個或一系列低分辨率圖像中重建高分辨率圖像的技術(shù)。該技術(shù)不僅可以增加圖像的高頻信息,還可以消除低分辨率。深度學(xué)習(xí)在現(xiàn)代數(shù)字圖像處理方面取得了突破性進(jìn)展。深層卷積神經(jīng)網(wǎng)絡(luò)通過大量的訓(xùn)練樣本學(xué)習(xí),獲取圖像中的相關(guān)信息,然后利用這些信息實現(xiàn)特定的功能。超分辨率（SR）圖像可以通過深度神經(jīng)網(wǎng)絡(luò)方法獲得,這些方法比以往所有傳統(tǒng)方法都能獲得更高的性能。在本研究中,我們提出了一種增強(qiáng)的深層卷積神經(jīng)網(wǎng)絡(luò),稱為... 

【文章頁數(shù)】：102 頁

【學(xué)位級別】：博士

【文章目錄】：
致謝
摘要
Abstract
縮寫和符號清單
術(shù)語表
1 Introduction
    1.1. Background of Research
    1.2. Significance of Research
    1.3. Problem Statement
    1.4. Research Content
        1.4.1. Super resolution
        1.4.2. Image Classification
        1.4.3. Change detection
    1.5. Research Objectives
    1.6. Main Contents and the Structure of the Thesis
2 Related Work-Different Deep learning Algorithms
    2.1 Introduction
    2.2 Existing Deep Learning Algorithms Applied in SR Images
        2.2.1 Sparse Coding Network
        2.2.2 Convolutional Neural Network(CNN)
        2.2.3 Deep Network Cascade(DNC)
        2.2.4 Restricted Boltzmann Machines(RBM)
        2.2.5 Deep Belief Network(DBN)
    2.3 Summary
3 Super-resolution Satellite images Based on Deep Learning
    3.1 Introduction
    3.2 Super-resolution Convolutional Neural Network Method
    3.3 Basic structure of deep convolutional neural network
    3.4 Proposed Enhancement Deep Convolution Neural Network(EDCNN)
        3.4.1 Feature extraction
        3.4.2 Detail prediction
        3.4.3 Reconstruction
    3.5 Quality masseurs
        3.5.1 PSNR
        3.5.2 SSIM
    3.6 Experiments, Results and the Performance analysis
        3.6.1 Experience Configurations
        3.6.2 Datasets and Study Area
        3.6.3 Performance Analysis
    3.7 Summary
4 Image classification based on Convolutional Neural Network
    4.1 Introduction
    4.2 Image Classification Methods
        4.2.1 Unsupervised Classification
        4.2.2 Supervised Classification
    4.3 Convolutional Neural Network Methods
        4.3.1 Image Classification based on CNN
        4.3.2 SegNet
    4.4 Applying Deep learning method in Classification process
        4.4.1 The acquisition of training data
        4.4.2 Convolutional neural networks(CNNs)
        4.4.3 Image Classification
    4.5 Experiments, Results and the Performance analysis
        4.5.1 Data resources
        4.5.2 Performance Analysis
        4.5.3 Enhance the performance
    4.6 Summary
5 SAR Images Change Detection based on Convolutional Neural Network
    5.1 Introduction
    5.2 Synthetic Aperture Radar (SAR)
        5.2.1 Synthetic aperture radar principle
        5.2.2 Synthetic Aperture Radar Features
    5.3 Applying deeplearning in Change Detection
    5.4 Our Proposed method
        5.4.1 Pre-classification and Sample Selection
        5.4.2 Convolutional Neural Networks
    5.5 Experiments, Results and the Performance analysis
        5.5.1 Datasets and Study Areas
        5.5.2 Performance Analysis
    5.6 Summary
6 Conclusion
    6.1 Overall
    6.2 Future Work
參考文獻(xiàn)
作者簡歷及在學(xué)研究成果
學(xué)位論文數(shù)據(jù)集


【參考文獻(xiàn)】：
期刊論文
[1]圖像超分辨率復(fù)原方法及應(yīng)用[J]. 陳健,高慧斌,王偉國,畢尋.  激光與光電子學(xué)進(jìn)展. 2015(02)



本文編號：3679511