Remote sensing of the Earth from space is changing our lives continuously. The weather forecasts now look impressively accurate! Agriculture also benefits from a more accurate monitoring of the Earth’s processes, and from a much closer look at the phenological periods, which allow farmers to improve their harvests. And what about the oceans? With new satellite sensors, we can now measure the salinity of the oceans and estimate their temperatures very precisely. All of this was unthinkable fifty years ago. Nowadays you may get all this information with a few clicks on your computer or smartphone.
Many satellites with several onboard sensors are currently flying over our heads and many are being built or planned for the next years. Each one has its own specificities, pushing further the boundaries of resolution in spatial, spectral or temporal terms. However, these advances imply an increased complexity, since the statistical characterization of remote sensing images turns out to be more difficult than in grayscale natural images. New problems must be faced, like the higher dimensionality of the pixels, the specific noise and uncertainty sources, the high spatial and spectral redundancy, and the inherently non-linear nature of the data structures. To make it even better, all these problems must be solved in different ways depending on the sensor and the acquisition process.
On the bright side, what looks like problems for the users also constitute research challenges with high potential for the communities involved in the processing of such information: the acquired signals have to be processed rapidly, transmitted, further corrected from different distortions, eventually compressed, and ultimately analyzed to extract valuable information. To do all of this better and better, the field of remote sensing research has become multidisciplinary. The traditional and physics-based concepts are now complemented with signal and image processing concepts, and this convergence creates a new field that is capable of managing the interface between the signal acquired and the physics of the surface of the Earth. Add to these computer vision and machine learning, and you get a whole lot of new exciting possibilities to understand and interpret the complex and extremely numerous images that satellites provide everyday.
Obviously, this is not the first book on remote sensing. There are many excellent treatises on this topic, but most of them are either specific to the application (land, atmosphere or oceans) or general-purpose textbooks (dealing with the physics of the involved processes). There are surprisingly few books dealing with the image processing part of the field and, due to the rapid evolution of the field, they are often outdated. This volume aims at filling this gap, by describing the main steps of the remote sensing information processing chain and building bridges between the remote sensing community and those who could impact it by sharing their own view of the image processing problem.
The book focuses on what we consider the most active, attractive and distinctive aspects of the field. It is organized in six chapters, which relate to other scientific communities (see Fig. 1). Chapter 1 offers a short review of the underlying physics, a list of the major applications, and a survey of the existing satellite platforms and sensors. We will present what remote sensing images look like, as well as the main preprocessing steps prior to proper image analysis. This chapter also provides pointers to companies, publication journals, conferences and main institutions involved in the field. Chapter 2, studies the statistics of hyperspectral images and compares them with those of natural gray-scale and color images. Some interesting conclusions are established: while spectral and spatial features certainly differ, the main statistical facts (spatial and spectral smoothness) are shared. This observation opens the door to new researchers from the classical image processing and computer vision communities. Chapter 3 deals with the problem of feature selection and extraction from remote sensing images: apart from the standard tools available in image/signal processing, we will see that specific physically-based features exist and need to be used to improve the representation, along with new methods specially designed to account for specific properties of the scenes. Chapter 4 tackles the important and very active fields of remote sensing image classification and target detection. Problems with describing the land use or detecting changes in multitemporal sequences are studied through segmentation, detection and adaptation methods. Researchers in this field have come up with many efficient algorithms, and these fields are perhaps the most influenced by the machine learning community. Chapter 5 presents the field of spectral unmixing analysis, which studies ways to retrieve the different materials contained in spatially coarse and spectrally mixed pixels. This relatively recent field shows interesting connections with blind source separation, sparse models, geometry, and manifold learning. Chapter 6 addresses one of the most active field in current remote sensing data analysis: the area of model inversion and parameter retrieval, which deals with the retrieval of biophysical quantities such as ocean salinity or chlorophyll content. This field is related to regression and function approximation, search in databases, regularization, and sparse learning.
Figure 1: Roadmap: The four main fields involved in remote sensing image processing and the most related chapters of this book.
After reading just these few pages, you can see that the field of remote sensing image processing is indeed very broad. For this reason, we could not cover all the interesting and relevant aspects exhaustively. For example, we do not cover radar or LiDAR signal processing, which are very active areas of research. Also, while we pay attention to the image features, the key issue of noise sources and corrections is not treated in detail. The same is true for the fields of image compression, super-resolution, co-registration, deconvolution or restoration to name a few. We preferred to be consistent with our presentation and leave room for a second volume!
The book is mainly targeted at researchers and graduate students in the fields of signal/image processing related to any aspect of color imaging, physics, computer vision and machine learning. However, we hope it will also find readers from other disciplines of science, and in particular in the fields of application where remote sensing images are used.
Gustavo Camps-Valls, Devis Tuia, Luis Gómez-Chova, Sandra Jiménez, and Jesús Malo València, 2011