Artificial Intelligence Applied to Satellite-based Remote Sensing Data for Earth Observation
- Length: 300 pages
- Edition: 1
- Language: English
- Publisher: The Institution of Engineering and Technology
- Publication Date: 2021-10-27
- ISBN-10: 1839532122
- ISBN-13: 9781839532122
- Sales Rank: #0 (See Top 100 Books)
Earth observation (EO) involves the collection, analysis, and presentation of data in order to monitor and assess the status and changes in natural and built environments. This technology has many applications including weather forecasting, tracking biodiversity, measuring land-use change, monitoring and responding to natural disasters, managing natural resources, monitoring emerging diseases and health risks, and predicting, adapting to and mitigating climate change.
This book shows how cutting-edge technologies such as artificial intelligence, including neural networks and deep learning, can be applied for processing satellite data for Earth observation. One of the objectives of this book is to explain how to develop a set of libraries for the implementation of artificial intelligence that could overcome some limits and encompass different aspects of research, ranging from data fusion to speckle filtering.
In the first part, the authors introduce remote sensing concepts and deep neural networks and convolutional neural networks. In the second part of the book, they present the main tools used for image processing, several simulations and the data processing of specific case studies as well as the testing of related datasets. The book ends with conclusions, open questions and future works and perspectives for artificial intelligence techniques applied to future satellite missions.
The book will be of interest to researchers focusing on using machine learning tools to process remote sensing data – particularly satellite data – for Earth observation. The book can also be used as a guide for researchers in many other fields of research who are interested in using ML techniques to process data and get reliable outcomes so they can make informed decisions for their specific objectives.
Cover Halftitle Page Series Page Title Page Copyright Contents Mathematical notation List of figures List of tables About the editors 1 The rise of Artificial Intelligence (AI) for Earth Observation (EO) 1.1 Embracing external trends 1.2 The Artificial Intelligence and Deep Learning revolution 1.3 Rise of Artificial Intelligence for Earth Observation 1.3.1 AI4EO data engine 1.3.2 AI@edge 1.4 Conclusions 2 Principles of satellite data analysis 2.1 Introduction 2.1.1 Principles underlying remote sensing 2.1.1.1 Electromagnetic radiation 2.1.1.2 Electromagnetic spectrum 2.1.2 Interaction between electromagnetic waves and matter 2.1.2.1 Reflection 2.1.2.2 Refraction 2.1.2.3 Diffraction 2.1.3 Other types of interaction 2.1.4 Interaction with the atmosphere 2.1.4.1 Scattering 2.1.4.2 Absorption 2.1.4.3 Refraction 2.2 Types of sensors 2.2.1 Active sensors 2.2.2 Passive sensors 2.3 Platforms 2.3.1 Ground-based platforms 2.3.2 Airborne platforms 2.3.3 Spaceborne platforms 2.4 Data characteristics 2.4.1 Operations for improving quality of data 2.5 Copernicus programme 2.5.1 The Sentinels 2.6 Focus on optical and SAR data 2.6.1 Principles of SAR 2.6.2 Classic processing workflows for SAR data 2.6.3 Principles of optical sensors 2.6.4 Classic processing workflows for optical data 2.6.5 Applications 2.6.5.1 An application of SAR data: the interferometry 2.6.5.2 An application of optical data: the object-based image analysis 2.7 Conclusions References 3 Artificial intelligence, machine learning and deep learning 3.1 Introduction 3.1.1 Historical pills on the birth of AI 3.2 Artificial intelligence, machine learning and deep learning 3.2.1 Machine learning 3.2.2 Deep learning 3.2.3 Real-life applications 3.3 Supervised, unsupervised and reinforcement learning 3.3.1 Supervised learning 3.3.2 Unsupervised learning 3.3.3 Reinforcement learning 3.4 Training and validation of a model 3.5 Cross-validation for supervised learning 3.6 Class imbalance and data augmentation 3.6.1 Class imbalance 3.6.2 Data augmentation 3.6.2.1 For tabular data 3.6.2.2 For image data 3.6.3 Noise robustness 3.6.4 Choice and use of a dataset 3.7 Applications 3.7.1 Remote sensing applications 3.7.1.1 Natural disasters 3.7.2 Other applications 3.7.2.1 Self-driving cars 3.7.2.2 E-commerce and social media 3.7.2.3 Space exploration 3.7.2.4 Fault detection of navigation systems 3.7.2.5 Medical applications 3.8 Conclusions References 4 Artificial neural network 4.1 Introduction 4.2 Artificial neural networks 4.2.1 Artificial neuron 4.3 Activation functions 4.3.1 Sigmoid function 4.3.2 Hyperbolic tangent function (tanh) 4.3.3 Rectified linear unit function 4.3.4 The bias 4.4 Forward and backward propagation 4.4.1 The gradient descent 4.4.2 Description 4.4.3 Stochastic gradient descent 4.4.4 Full batch gradient descent 4.4.5 Mini-batch gradient descent 4.4.5.1 Tips 4.4.6 Batch normalisation 4.4.7 The vanishing gradient 4.4.7.1 Solutions 4.4.8 The gradient explosion 4.4.8.1 Solution 4.4.9 The back propagation 4.4.10 A numerical example 4.5 Overfitting and underfitting 4.5.1 Bias and variance 4.5.2 Early stopping of training 4.5.3 Dropout 4.6 Regularisation techniques 4.6.1 Parameter norm penalties 4.6.1.1 Regularisation L 4.6.1.2 Regularisation L 4.7 Conclusions References 5 Convolutional neural networks 5.1 Introduction 5.2 Convolutional neural networks 5.2.1 Motivation 5.2.2 The convolution operation 5.2.3 Parameter sharing and sparse connectivity 5.3 Networks and layers structure 5.4 Kernels and feature maps 5.4.1 Stride 5.4.2 Multidimensional input 5.4.3 Padding 5.5 ReLU layer 5.6 Pooling 5.7 Batch normalisation 5.7.1 Motivation 5.7.2 How does batch normalisation work? 5.8 Common tasks performed by CNNs 5.9 Conclusions References 6 How to create a proper EO dataset 6.1 Introduction 6.2 Non-automated method 6.2.1 The open access hub 6.2.2 SNAP – Sentinel Application Platform 6.3 Automated method 6.3.1 Google Earth Engine – online editor 6.3.2 Google Earth Engine – Python API 6.4 Dataset annotation 6.5 Training dataset for image classification tasks 6.5.1 Annotation tool for image classification – makesense.ai 6.6 Training dataset for semantic segmentation tasks 6.6.1 Annotation tool for semantic segmentation – Google Earth Engine online editor 6.7 Training dataset for object detection tasks 6.7.1 Annotation tool for semantic segmentation – LabelImg 6.8 Conclusions References 7 How to develop your network with Python and Keras 7.1 Introduction 7.1.1 TensorFlow and Keras 7.1.2 Tensors 7.2 Google Colaboratory 7.2.1 How to work with GPU? 7.2.2 How to connect Google Drive? 7.3 Examples 7.3.1 Noise filtering 7.3.1.1 Step 1 – Define a generator for the dataset 7.3.1.2 Step 2 – Build a model 7.3.1.3 Step 3 – Train and test the model 7.3.2 Classification 7.3.2.1 Step 1 – Define a generator for the dataset 7.3.2.2 Step 2 – Build a model 7.3.2.3 Step 3 – Train and test the model 7.3.3 Detection 7.3.3.1 Step 1 – Define a generator for the dataset 7.3.3.2 Step 2 – Build a model 7.3.3.3 Step 3 – Train and test the model 7.4 Conclusions References 8 A classification problem 8.1 Introduction 8.2 Case study A: volcanic eruptions binary classification 8.2.1 The catalog 8.2.2 Data preparation and manipulation 8.2.3 Dataset expansion 8.2.4 Proposed model 8.2.4.1 Image Loader 8.2.5 Training 8.2.6 Test on test dataset 8.2.7 Future works 8.3 Case Study B: Landslides binary classification 8.3.1 Landslide classification 8.3.2 The CNN for landslides binary classification 8.3.3 The method based on optical images 8.3.4 SAR data and creation of the dataset 8.3.4.1 GRD and SLC SAR data 8.3.5 The proposed CNN for SAR data 8.3.6 Results 8.3.7 Future developments 8.3.7.1 Dataset maintenance 8.3.7.2 RGB composite with SLC data 8.4 Case Study C: Crops-type classification 8.4.1 CTC problem 8.4.2 Dataset 8.4.3 Classification using CNNs and transfer learning 8.4.4 Results 8.4.5 Future works 8.5 Case Study D: Segmentation 8.5.1 Segmentation tool and algorithms 8.5.2 Experimental analysis and results 8.5.2.1 Contrast split segmentation 8.5.2.2 Multiresolution segmentation 8.5.3 Integration of deep learning and object-based analysis 8.6 Conclusions References 9 A generation problem 9.1 Introduction 9.2 Deep generative models 9.2.1 Variational autoencoders 9.2.2 Generative adversarial networks 9.3 Case study: Generation of optical-like images 9.3.1 Dataset and proposed methods 9.4 Model for the generation of optical-like data 9.4.1 Generator 9.4.2 Discriminator 9.4.3 Loss function selection 9.5 Results 9.5.1 Test on simulated Sentinel-1 images 9.5.2 Test on real Sentinel-1 images 9.6 Conclusions References 10 A filtering problem: SAR speckle filtering 10.1 Introduction 10.1.1 Speckle filtering motivation 10.1.2 Speckle reduction 10.1.3 Speckle filtering examples 10.2 Case study: Sentinel-1 speckle filtering 10.2.1 Multi-modal speckle filtering 10.2.2 Pure SAR strategy 10.2.3 CNN model for SAR speckle filtering 10.2.4 Network architecture 10.2.5 Loss function selection 10.3 Evaluation metrics 10.3.1 Peak signal-to-noise ratio 10.3.2 The structural similarity index 10.4 Results 10.4.1 Data preparation 10.4.2 Test on small patches of a Sentinel-1 image 10.4.2.1 Test on the training dataset 10.4.2.2 Test on validation dataset 10.4.2.3 Comparison with the state of the art 10.4.3 Test on a whole Sentinel-1 image 10.5 Conclusions References 11 Future perspectives and conclusions 11.1 Introduction 11.2 A first prototype 11.2.1 Raspberry Pi 11.2.2 Movidius Stick 11.2.3 Camera 11.2.4 Proposed model 11.2.5 Results 11.2.6 Implementation on Raspberry and Movidius Stick 11.2.7 OpenVINO library 11.3 Future satellite missions with AI on board 11.3.1 ESA first intelligent satellites: φ -Sat 1 and 11.4 Conclusions 11.5 Acknowledgements References Index Back Cover
Donate to keep this site alive
1. Disable the AdBlock plugin. Otherwise, you may not get any links.
2. Solve the CAPTCHA.
3. Click download link.
4. Lead to download server to download.