Change Detection and Image Time Series Analysis 2: Supervised Methods

Change Detection and Image Time Series Analysis 2 presents supervised machine-learning-based methods for temporal evolution analysis by using image time series associated with Earth observation data. Chapter 1 addresses the fusion of multisensor, multiresolution and multitemporal data. It proposes two supervised solutions that are based on a Markov random field: the first relies on a quad-tree and the second is specifically designed to deal with multimission, multifrequency and multiresolution time series. Chapter 2 provides an overview of pixel based methods for time series classification, from the earliest shallow learning methods to the most recent deep-learning-based approaches. Chapter 3 focuses on very high spatial resolution data time series and on the use of semantic information for modeling spatio-temporal evolution patterns. Chapter 4 centers on the challenges of dense time series analysis, including pre processing aspects and a taxonomy of existing methodologies. Finally, since the evaluation of a learning system can be subject to multiple considerations, Chapters 5 and 6 offer extensive evaluations of the methodologies and learning frameworks used to produce change maps, in the context of multiclass and/or multilabel change classification issues.

Table of Contents:
Contents Preface ix Abdourrahmane M. ATTO, Francesca BOVOLO and Lorenzo BRUZZONE List of Notations Chapter 1 Hierarchical Markov Random Fields for High Resolution Land Cover Classification of Multisensor and Multiresolution Image Time Series 1 Ihsen HEDHLI, Gabriele MOSER, Sebastiano B. SERPICO and Josiane ZERUBIA 1.1. Introduction 1 1.1.1. The role of multisensor data in time series classification 1 1.1.2. Multisensor and multiresolution classification 2 1.1.3.Previouswork 5 1.2. Methodology 9 1.2.1. Overview of the proposed approaches 9 1.2.2. Hierarchical model associated with the first proposed method 10 1.2.3. Hierarchical model associated with the second proposed method 13 1.2.4. Multisensor hierarchical MPM inference 14 1.2.5. Probability density estimation through finite mixtures 17 1.3.Examplesofexperimentalresults 19 1.3.1.Resultsofthefirstmethod 19 1.3.2.Resultsofthesecondmethod 22 1.4.Conclusion 26 xiii 1.5.Acknowledgments 26 1.6.References 27 Chapter 2 Pixel-based Classification Techniques for Satellite Image Time Series 33 Charlotte PELLETIER and Silvia VALERO 2.1. Introduction 33 2.2. Basic concepts in supervised remote sensing classification 35 2.2.1. Preparing data before it is fed into classification algorithms 35 2.2.2. Key considerations when training supervised classifiers 39 2.2.3. Performance evaluation of supervised classifiers 41 2.3.Traditionalclassificationalgorithms 45 2.3.1. Support vector machines 45 2.3.2. Random forests 51 2.3.3. k-nearest neighbor 56 2.4. Classification strategies based on temporal feature representations 59 2.4.1. Phenology-based classification approaches 60 2.4.2 Dictionary-based classificationapproaches 61 2.4.3 Shapelet-based classificationapproaches 62 2.5.Deeplearningapproaches 63 2.5.1. Introduction to deep learning 64 2.5.2.Convolutionalneuralnetworks 68 2.5.3.Recurrentneuralnetworks 71 2.6.References 75 Chapter 3 Semantic Analysis of Satellite Image Time Series 85 Corneliu Octavian DUMITRU and Mihai DATCU 3.1. Introduction 85 3.1.1.TypicalSITSexamples 89 3.1.2. Irregular acquisitions 90 3.1.3.Thechapterstructure 96 3.2.WhyaresemanticsneededinSITS? 96 3.3.Similaritymetrics 97 3.4. Feature methods 98 3.5. Classification methods 98 3.5.1.Activelearning 99 3.5.2.Relevancefeedback 100 3.5.3. Compression-based pattern recognition 100 3.5.4.LatentDirichletallocation 101 3.6.Conclusion 102 vii 3.7.Acknowledgments 105 3.8.References 105 Chapter 4 Optical Satellite Image Time Series Analysis for Environment Applications: From Classical Methods to Deep Learning and Beyond 109 Matthieu MOLINIER, Jukka MIETTINEN,DinoIENCO,ShiQIU and Zhe ZHU 4.1. Introduction 109 4.2. Annual time series 111 4.2.1. Overview of annual time series methods 111 4.2.2 Examples of annual times series analysis applications for environmentalmonitoring 112 4.2.3.Towardsdensetimeseriesanalysis 116 4.3. Dense time series analysis using all available data 117 4.3.1. Making dense time series consistent 118 4.3.2. Change detection methods 121 4.3.3.Summaryandfuturedevelopments 125 4.4. Deep learning-based time series analysis approaches 126 4.4.1 Recurrent Neural Network (RNN) for Satellite Image TimeSeries 129 4.4.2 Convolutional Neural Networks (CNN) for Satellite Image TimeSeries 131 4.4.3. Hybrid models: Convolutional Recurrent Neural Network (ConvRNN) models for Satellite Image Time Series 134 4.4.4. Synthesis and future developments 136 4.5. Beyond satellite image time series and deep learning: convergence between time series and video approaches 136 4.5.1 Increased image acquisition frequency: from time series to spacebornetime-lapseandvideos 137 4.5.2. Deep learning and computer vision as technology enablers 138 4.5.3.Futuresteps 139 4.6.References 140 Chapter 5 A Review on Multi-temporal Earthquake Damage Assessment Using Satellite Images 155 Gülşen TAŞKIN, EsraERTEN and Enes Oğuzhan ALATAŞ 5.1. Introduction 155 5.1.1. Research methodology and statistics 159 5.2. Satellite-based earthquake damage assessment 165 5.3. Pre-processing of satellite images before damage assessment 167 5.4. Multi-source image analysis 168 5.5. Contextual feature mining for damage assessment 169 5.5.1.Texturalfeatures 170 5.5.2. Filter-based methods 173 5.6. Multi-temporal image analysis for damage assessment 175 5.6.1. Use of machine learning in damage assessment problem 176 5.6.2. Rapid earthquake damage assessment 180 5.7. Understanding damage following an earthquake using satellite-based SAR 181 5.7.1. SAR fundamental parameters and acquisition vector 185 5.7.2. Coherent methods for damage assessment 188 5.7.3. Incoherent methods for damage assessment 192 5.7.4. Post-earthquake-only SAR data-based damage assessment 195 5.7.5 Combination of coherent and incoherent methods for damage assessment 196 5.7.6.Summary 198 5.8. Use of auxiliary data sources 200 5.9.Damagegrades 200 5.10.Conclusionanddiscussion 203 5.11.References 205 Chapter 6 Multiclass Multilabel Change of State Transfer Learning from Image Time Series 223 Abdourrahmane M. ATTO,HélaHADHRI, FlavienVERNIER and Emmanuel TROUVÉ 6.1. Introduction 223 6.2. Coarse- to fine-grained change of state dataset 225 6.3. Deep transfer learning models for change of state classification 232 6.3.1.Deeplearningmodellibrary 232 6.3.2.GraphstructuresfortheCNNlibrary 234 6.3.3. Dimensionalities of the learnables for the CNN library 236 6.4.Changeofstateanalysis 237 6.4.1 Transfer learning adaptations for the change of state classificationissues 238 6.4.2.Experimentalresults 239 6.5.Conclusion 243 6.6.Acknowledgments 244 6.7.References 244 List of Authors 247 Index 249 Summary of Volume 1 253

About the Author :
Abdourrahmane M. Atto is Associate Professor at the University Savoie Mont Blanc, France. His research interests include mathematical methods and models for artificial intelligence and image time series. Francesca Bovolo is the Head of the Remote Sensing for Digital Earth Unit, Fondazione Bruno Kessler, Italy. Her research interests include remote sensing image time series analysis, content-based time series retrieval and radar sounders. Lorenzo Bruzzone is Professor of Telecommunications and the Founder and Director of the Remote Sensing Laboratory at the University of Trento, Italy. His research interests include remote sensing, machine learning and pattern recognition.