This book elaborates fuzzy Machine Learning and Deep Learning models for single class mapping from multi-sensor, multi-temporal remote sensing images while handling mixed pixels and noise. It also covers the ways of pre-processing and spectral dimensionality reduction of temporal data. Further, it discusses the ‘individual sample as mean’ training approach to handle heterogeneity within a class. The appendix section of the book includes case studies such as mapping crop type, forest species, and stubble burnt paddy fields.
There are numerous Machine Learning models available for use in a variety of applications. Machine Learning classifers can fall into statistical categories such as c-Means, maximum-likelihood classifers, and decision-tree categories, such as random forest and Classifcation and Regression Tree (CART). Fuzzy c-Means (FCM), Possibilistic c-Means (PCM), Noise Clustering (NC), and Modifed Possibilistic c-Means (MPCM) are examples of fuzzy-logic algorithms. The Machine Learning model should be chosen in such a way that it can deal with mixed pixels, non-linearity between classes, and noisy pixels. Machine Learning models should be able to map a single class of interest, which has a wide range of applications, where only one class of interest needs to be mapped from remote-sensing data. Machine Learning models should also be able to deal with heterogeneity within classes, which is caused by the fact that training data for each class is not homogeneous. So, in this section, fuzzy classifers are discussed, which addresses the raised concerns. In this chapter, mathematical formulas and algorithms of the fuzzy-based algorithms have been explained. The description starts with Fuzzy c-Means (FCM), moves on to Possibilistic c-Means (PCM) and Noise Clustering, and then Modifed Possibilistic c-Means (MPCM) is described.
Key features:
Focuses on use of multi-sensor, multi-temporal data while handling spectral overlap between classes
Discusses range of fuzzy/deep learning models capable to extract specific single class and separates noise
Describes pre-processing while using spectral, textural, CBSI indices, and back scatter coefficient/Radar Vegetation Index (RVI)
Discusses the role of training data to handle the heterogeneity within a class
Supports multi-sensor and multi-temporal data processing through in-house SMIC software
Includes case studies and practical applications for single class mapping
This book is intended for graduate/postgraduate students, research scholars, and professionals working in environmental, geography, Computer Sciences, remote sensing, geoinformatics, forestry, agriculture, post-disaster, urban transition studies, and other related areas.
Author(s): Anil Kumar, Priyadarshi Upadhyay, Uttara Singh
Publisher: CRC Press
Year: 2023
Language: English
Pages: 178
Cover
Half Title
Title
Copyright
Dedication
Contents
Foreword
Preface
Our Gratitude with three Rs
Author Biographies
List of Abbreviations
Chapter 1 Remote-Sensing Images
1.1 Introduction
1.2 Introduction to Multispectral Remote-Sensing
1.3 Introduction to Hyperspectral Remote-Sensing
1.3.1 Hyperspectral Data Pre-processing
1.3.2 Endmember Extraction
1.4 Introduction to SAR Remote-Sensing
1.5 Dimensionality Reduction
1.6 Summary
Bibliography
Chapter 2 Evolution of Pixel-Based Spectral Indices
2.1 Introduction
2.2 Spatial Information
2.3 Spectral Indices
2.4 Texture-Based Spatial Indices
2.5 Summary
Bibliography
Chapter 3 Multi-Sensor, Multi-Temporal Remote-Sensing
3.1 Introduction
3.2 Temporal Vegetation Indices
3.3 Specific Single Class Mapping
3.4 Indices for Temporal Data
3.5 Temporal Data With Multi-Sensor Concept
3.6 Summary
Bibliography
Chapter 4 Training Approaches—Role of Training Data
4.1 Introduction
4.2 Handling Heterogeneity Within a Class
4.3 Manual or Region Growing Method for Training-Samples Collection
4.4 Extension of Training Samples
4.5 Cognitive Approach to Train Classifier
4.6 Specific Class Mapping Applications
4.7 Summary
Bibliography
Chapter 5 Machine-Learning Models for Specific-Class Mapping
5.1 Introduction
5.2 Fuzzy Set-Theory-Based Algorithms
5.3 Fuzzy c-Means (FCM) Algorithm
5.4 Possibilistic c-Means Classification
5.5 Noise Clustering
5.6 Modified Possibilistic c-Means (MPCM) Algorithm
5.7 Summary
Bibliography
Chapter 6 Learning-Based Algorithms for Specific-Class Mapping
6.1 Introduction
6.2 Convolutional Neural Networks (CNN)
6.3 Recurrent Neural Networks (RNN)
6.4 Difference Between RNN and CNN
6.5 Long Short-Term Memory (LSTM)
6.6 Gated Recurrent Unit (GRU)
6.7 Difference Between GRU & LSTM
6.8 Summary
Bibliography
Appendix A1 Specific Single Class Mapping Case Studies
A1. Fuzzy Versus Deep-Learning Classifiers for Transplanted Paddy Fields Mapping
A2. Dual-Sensor Temporal Data for Mapping Forest Vegetation Species and Specific-Crop Mapping
A3. Handling Heterogeneity With Training Samples Using Individual-Sample-as-Mean Approach for Isabgol (Psyllium Husk) Medicinal Crop
A4. Sunflower Crop Mapping Using Fuzzy Classification While Studying Effect of Red-Edge Bands
A5. Mapping Burnt Paddy Fields Using Two Dates’ Temporal Sentinel-2 Data
A6. Mapping Ten-Year-Old Dalbergia Sissoo Forest Species
A7. Transition Building Footprints Mapping
Appendix A2 SMIC—Temporal Data-Processing Module for Specific-Class Mapping
Index
