Advantages and Pitfalls of Pattern Recognition: Selected Cases in Geophysics: Volume 3

Description

Advantages and Pitfalls of Pattern Recognition presents various methods of pattern recognition and classification, useful to geophysicists, geochemists, geologists, geographers, data analysts, and educators and students of geosciences. Scientific and technological progress has dramatically improved the knowledge of our planet with huge amounts of digital data available in various fields of Earth Sciences, such as geology, geophysics, and geography. This has led to a new perspective of data analysis, requiring specific techniques that take several features into consideration rather than single parameters. Pattern recognition techniques offer a suitable key for processing and extracting useful information from the data of multivariate analysis. This book explores both supervised and unsupervised pattern recognition techniques, while providing insight into their application. Cover Advantages and Pitfalls of Pattern Recognition: Selected Cases in Geophysics Copyright Preface Acknowledgments Part I 1. Patterns, objects, and features 1.1 Objects and patterns 1.2 Features 1.2.1 Types 1.2.2 Feature vectors 1.2.3 Feature extraction 1.2.3.1 Delineating segments 1.2.3.2 Delineating regions 1.2.4 Transformations 1.2.4.1 Karhunen–Loève transformation (Principal Component Analysis) 1.2.4.2 Independent Component Analysis 1.2.4.3 Fourier transform 1.2.4.4 Short-time Fourier transform and spectrograms 1.2.4.5 Discrete wavelet transforms 1.2.5 Standardization, normalization, and other preprocessing steps 1.2.5.1 Comments 1.2.5.2 Outlier removal 1.2.5.3 Missing data 1.2.6 Curse of dimensionality 1.2.7 Feature selection Appendix 1 Basic notions on statistics A1.1 Statistical parameters of an ensemble A1.2 Distinction of ensembles 2. Supervised learning 2.1 Introduction 2.2 Discriminant analysis 2.2.1 Test ban treaty—some history 2.2.2 The MS–mb criterion for nuclear test identification 2.2.3 Linear Discriminant Analysis 2.3 The linear perceptron 2.4 Solving the XOR problem: classification using multilayer perceptrons (MLPs) 2.4.1 Nonlinear perceptrons 2.5 Support vector machines (SVMs) 2.5.1 Linear SVM 2.5.2 Nonlinear SVM, kernels 2.6 Hidden Markov Models (HMMs)/sequential data 2.6.1 Background—from patterns and classes to sequences and processes 2.6.2 The three problems of HMMs 2.6.3 Including prior knowledge/model dimensions and topology 2.6.4 Extension to conditional random fields 2.7 Bayesian networks Appendix 2 Appendix 2.1 Fisher's linear discriminant analysis Appendix 2.2 The perceptron Backpropagation Appendix 2.3 SVM optimization of the margins Appendix 2.4. Hidden Markov models Appendix 2.4.1. Evaluation Appendix 2.4.2. Decoding—the Viterbi algorithm Appendix 2.4.3. Training—the expectation–maximization /Baum–Welch algorithm 3. Unsupervised learning 3.1 Introduction 3.1.1 Metrics of (dis)similarity 3.1.2 Clustering 3.1.2.1 Partitioning clustering 3.1.2.1.1 Fuzzy clustering 3.1.2.2 Hierarchical clustering 3.1.2.3 Density-based clustering 3.2 Self-Organizing Maps 3.2.1 Training of an SOM Appendix 3 Appendix 3.1. Analysis of variance (ANOVA) Appendix 3.2 Minimum distance property for the determinant criterion Appendix 3.3. SOM quality Topological error Designing the map Part II 4. Applications of supervised learning 4.1 Introduction 4.2 Classification of seismic waveforms recorded on volcanoes 4.2.1 Signal classification of explosion quakes at Stromboli 4.2.2 Cross-validation issues 4.3 Infrasound classification 4.3.1 Infrasound monitoring at Mt Etna—classification with SVM 4.4 SVM classification of rocks 4.5 Inversion with MLP 4.5.1 Identification of parameters governing seismic waveforms 4.5.2 Integrated inversion of geophysical data 4.6 MLP in regression and interpolation 4.7 Regression with SVM 4.7.1 Background 4.7.2 Brief considerations on pros and cons of SVM and MLP in regression problems 4.8 Classification by hidden Markov models and dynamic Bayesian networks: application to seismic waveforms of tectonic, volcani ... 4.8.1 Background 4.8.2 Signals related to volcanic and tectonic activity 4.8.3 Classification of icequake and nonterrestrial seismic waveforms as base for further research —HMM 4.8.3.1 Icequakes 4.8.3.2 Moon quakes 4.8.3.3 Classification of seismic waveforms using dynamic Bayesian networks 4.9 Natural hazard analyses—HMMs and BNs 4.9.1 Estimating volcanic unrest 4.9.2 Reasoning under uncertainty—tsunami early warning tasks Appendix 4.1. Normalization issues Appendix 4.2. SVM Regression Appendix 4.3. Bias—Variance Trade-off in Curve Fitting 5. Applications with unsupervised learning 5.1 Introduction 5.2 Cluster analysis of volcanic tremor data 5.3 Density based clustering 5.4 Climate zones 5.5 Monitoring spectral characteristics of seismic signals and volcano alert 5.6 Directional features Appendix 5 Appendix 5.1 Davies-Bouldin index Appendix 5.2 Dunn index Appendix 5.3 Silhouette index Appendix 5.4 Gap index Appendix 5.5 Variation of information Part III 6. A posteriori analyses—advantages and pitfalls of pattern recognition techniques 6.1 Introduction 6.2 Testing issues 6.3 Measuring error 6.4 Targets 6.5 Objects 6.6 Features and metrics 6.7 Concluding remarks 6.7.1 Multilayer perceptrons 6.7.2 Support Vector Machines 6.7.3 MLP and SVM in regression analysis 6.7.4 Hidden Markov models and Bayesian networks 6.7.5 Supervised and unsupervised learning 7. Software manuals 7.1 Example scripts related to Chapter 2 7.1.1 Linear discrimination, principal components, and marginal distributions 7.1.2 The perceptron 7.1.3 Support Vector Machines 7.1.4 HMM example routines (from Theodoridis et al., 2010, see http://booksite.elsevier.com/9780123744869) 7.2 Example scripts and programs related to Chapter 3 (unsupervised learning) 7.2.1 K-means clustering 7.2.2 Mixed models 7.2.3 Expectation maximization clusters 7.2.4 Fuzzy clustering 7.2.5 Hierarchical clustering 7.2.6 Density-based clustering 7.2.7 Unsupervised learning toolbox: KKAnalysis 7.2.7.1 Preliminaries 7.2.7.2 Installation 7.2.7.3 Files 7.2.7.3.1 Input files 7.2.7.3.2 Output files 7.2.7.4 Getting started 7.2.7.4.1 The “Input File” frame 7.2.7.4.2 The “figures” frame 7.2.7.5 Configuring KKAnalysis—the “settings” 7.3 Programs related to applications (Chapter 4) 7.3.1 Back propagation neural network (BPNN) 7.3.2 SVM library 7.4 Miscellaneous 7.4.1 DMGA—generating ground deformation, magnetic and gravity data 7.4.2 Treating fault plane solution data Bibliography Bibliography Index A B C D E F G H I J K L M N O P Q R S T U V W X Back Cover