Signal and Image Processing in Hyper Spectral Remote Sensing

Aims And Scope: Hyper spectral imaging has recently emerged as one of the very promising technologies in remote sensing, enabling applications that may have been previously seen as impossible in multi spectral imaging. Hyper spectral cameras deployed in current airborne or satellite systems can cover the visible and nearinfrared wavelengths at a resolution of 10nm, with more than 200 spectral channels. This vastly increased spectral information content creates a unique opportunity for numerous applications, such as mineral identification, agriculture, environment monitoring, terrain classification, object detection, change detection, and many more. Hyper spectral imaging is also a key technique for planetary exploration, astrophysics, and non-remote sensing problems such as food inspection and forensics. Remarkably, these meaningful and important applications have led to a wide variety of signal processing problems, which have attracted growing attention and contributions from the signal processing, image processing and machine learning communities. In particular, we have witnessed developments that are far from being just a straight application of a signal processing technique. Instead, some of them turn out to provide new insights and open new dimensions for fundamental signal processing research. For example, it has recently become clear that the unmixing topic in hyper spectral remote sensing has formed a new branch of blind source separation techniques, wherein the exploitation of special source characteristics, such as local sparsity, has been found to provide very effective blind separation solutions. The same goes with the classification and detection topics, where the utilization of contextual information or combined spatial-spectral processing has resulted in new paradigms. Moreover, the recent research trend indicates that hyper spectral signal and image processing is embracing frontier signal processing concepts very quickly-this includes sparse signal processing, compressive sensing, and convex and non-convex optimization, just to name a few. The aim of this special issue is to gather high-quality tutorial-style articles that introduce key signal processing topics arising from hyperspectral remote sensing, demonstrate the insight and uniqueness of signal processing techniques established in this area, and/or provide overviews of the latest trends. In particular, we wish to shift the perspective from the remote sensing side to signal processing, and extract insight behind the signal processing developments happening in hyper spectral remote sensing. While this is the focus of this special issue, we may also

welcome application-oriented papers that can tell a good story regarding how signal processing makes a difference. Words includes remote sensing: Un-mixing, both linear and nonlinear, and both semi-supervised and unsupervised classification target or anomaly detection at a sub-pixel level coded aperture and compressive sensing sparse signal processing, which includes sparse regression, dictionary learning, multiple measurement vector models, etc convex and non-convex optimization contextual information or combined spatial-spectral processing Bayesian and statistical signal processing nonlinear manifold learning, graph theoretic methods dimension reduction, subspace identification, non-negative matrix factorization Introduction and Outline: A hyper spectral imager records a spectrum of the light from each pixel in the scene. Hyper spectral image analysis exploits this extra spectral information For an open land scene: Total spectrum for a pixel is a weighted sum of the spectra of what is in that pixel The imager and method of exploitation must be tailored to the scene and the desired products. Milestones in the history of Remote Sensing: Remote sensing began in 1840 when balloonists used new camera technology to take pictures. At the turn of the century there was a pigeon fleet in Europe. Camera systems Notes placed on V-2 were rockets tested at White Sands, NM after WW II. Sputnik in 1957 changed our outlook toward using outer space as a place from which to observe the earth. 1800: Discovery of infrared by Sir William Herschel 1839: Beginning of practice of photography

1847: Infrared spectrum shown by A. H. L. Fizeau & J.B.L. Foucault to share properties with visible light 1850-1860: Photography from balloons 1873: Theory of electromagnetic energy developed by James Clerk Maxwell 1909: Photography from airplanes 1914-1918: World War 1: aerial reconnaissance 1920-1930: Development & initial application of aerial photography & photogrammetry 1929-1939: Economic depression generates environmental crises that lead to governmental application of aerial photography. 1930-1940: Development of radars in Germany, US & UK 1839-1945: World War II: application of non-visible portions of electromagnetic spectrum, training of persons in acquisition and interpretation of air-photos. 1950-1960: Military research and development 1850-1860: Photography from balloons 1956 : Col-well research on plant disease detection with infrared photography. 1960-1970: First use of term remote sensing, start TIROS weather satellite, Skylab 1972: Launch of Landsat 1 1970-1980: Rapid advances in digital image processing 1980-1990: Landsat 4: new generation of Landsat sensors 1986: SPOT French Earth Observation satellite 1980: Development of hyper spectral sensor What is REMOTE SENSING ? Definition of Remote Sensing "Remote sensing is the science (and to some extent, art) of acquiring information about the Earth's surface without actually being in contact with it. This is done by sensing and recording reflected or emitted energy and processing, analyzing, and applying that information" (from Canadian Center for Remote Sensing). REMOTE SENSING includes all methods and techniques used to gain qualitative and quantitative information about distant objects without coming into direct contact with these objects. example: Look-Look,NO-Touch

Multiple light paths Scattering due to: atmosphere, aerosols, water surface,suspended particles bottom Absorption due to: atmosphere, aerosols,suspended particles,dissolved matter,Scattering and absorption are convolved Physical and biological modeling of the scene is often required to analyze the hyper spectral image.

Introduction to Hyper spectral Imaging:

A hyper spectral imager records a spectrum of the light from each pixel in the scene Hyper spectral image analysis exploits this extra spectral information For an open land scene:

Spectral Decomposition for an open land scene

Total spectrum for a pixel is a what is in that pixel

The imager and method of exploitation must be tailored to the scene and the desired products.

HICO based on experience with PHILLS:

PHILLS image of shallow water features near Lee Stocking Island, Bahamas used to develop and validate hyper spectral algorithms for bathymetry, bottom type and water clarity. Airborne Experiments with the Portable Hyperspectral Imager for LowLight Spectroscopy (PHILLS) demonstrated: Sensor design. Processing algorithms. Shallow water bathymetry, hazards to navigation, and beach traffic ability from hyper spectral remote sensing data.

What is the Hyperspectral Imager for the Coastal Ocean (HICO)?

HICO is an experiment to see what we gain by imaging the coastal ocean at higher resolution from space. The HICO sensor: first space borne imaging spectrometer for coastal oceans samples coastal regions at <100 m (400 to 900 nm: at 5.7 nm) high signal-to-noise ratio to resolve the complexity of the coastal ocean Sponsored as an Innovative Naval Prototype (INP) by the Office of Naval Research: Goal to reduce cost and a greatly shortened schedule. Start of Project to Sensor Delivery in 16 months Launched to the ISS September 10, 2009

HICO image of Hong Kong, October 2, 2009. HICO is integrated and flown under the direction of DoDs Space Test Program

HICO meets Performance Requirements: