Robert A. Neville

Spectral unmixing of hyperspectral imagery for mineral exploration: comparison of results from SFSI and AVIRIS

Hyperspectral image data sets acquired near Cuprite, Nevada, in 1995 with the Short-Wave Infrared (SWIR) Full Spectrum Imager (SFSI) and in 1996 with the Airborne Visible-Infrared Imaging Spectrometer (AVIRIS) are analysed with a spectral unmixing procedure and the results compared. The nominal pixel centre spacings are 1.0 by 1.5 m for SFSI and 16.2 by 18.1 m for AVIRIS across track and along track, respectively; the region imaged by SFSI is a small portion of the full AVIRIS scene. Both data cubes have nominal spectral band centre spacings of approximately 10 nm. The image data, converted to radiance units, are atmospherically corrected and converted to surface reflectances. Spectral end members are extracted automatically from the two data sets; those representing mineral species common to both are compared to each other and to reference spectra obtained with a field instrument, the Portable Infrared Mineral Analyser (PIMA). The full sets of end members are used in a constrained linear unmixing of the respective hyperspectral image cubes. The resulting unmixing fraction images derived from the AVIRIS and SFSI data sets for the minerals alunite, buddingtonite, kaolinite, and opal correlate well, with correlation coefficients ranging from 0.75 to 0.91, after compensation for shadowing and misregistration effects.

Preprocessing of EO-1 Hyperion data

A procedure for processing hyperspectral data acquired with Hyperion has been developed with an aim to correct for sensor artifacts and atmospheric and geometric effects. Advances in preprocessing of hyperspectral remote sensing data have enabled more accurate atmospheric correction and have led to the development of new information extraction techniques in the areas of agriculture, forestry, geosciences, and environmental monitoring. These processing and analysis tools have been incorporated into Imaging Spectrometer Data Analysis Systems (ISDAS), a software package developed at the Canada Centre for Remote Sensing (CCRS). The procedure, as applied for Hyperion data, begins with geometric corrections to the short-wave infrared (SWIR) component to register the SWIR and visible near-infrared (VNIR) data spatially. This is followed by the removal of stripes and pixel (column) dropouts and noise reduction, using recently developed automated software tools. The data cube is subsequently analyzed using keystone and spectral smile detection software to characterize these distortions. Included in the smile detection procedure is an optional gain and offset correction technique. The radiance data are converted to reflectance using a MODTRAN-based atmospheric correction procedure. Only at this point are the data corrected for smile effects. Any artifacts still remaining after these corrections are removed by post-processing.

Automatic destriping of Hyperion imagery based on spectral moment matching

As a pushbroom imaging spectrometer, the Hyperion sensor records each along-track image column with a single detector element in one of its area detector arrays. This image acquisition configuration is prone to exhibiting along-track striping artefacts, if the instrument has not undergone a recent, proper uniformity calibration. These defective columns must be corrected before performing further processing and quantitative analyses. Spatial moment matching (SpaMM) is currently the most commonly used method for destriping Hyperion imagery. However, since SpaMM was originally developed for dealing with remotely sensed multispectral data, the abundant spectral information embedded in hyperspectral data is not fully utilized in the method. This paper proposes a new methodology that can automatically and more accurately remove the striping artefacts from Hyperion imagery. The technique is called spectral moment matching (SpcMM) because it uses spectral autocorrelation instead of spatial autocorrelation to estimate the expected mean and standard deviation of a subscene, which is comprised of the image pixels acquired by an identical detector element (an along-track image column in the case of Hyperion data). The basis of the algorithm is the observation that there are usually highly correlated groups of bands in a hyperspectral image cube; the statistics of the subscenes measured by the corresponding detectors in a set of highly correlated bands are usually very similar. The possibility of introducing undesired side effects into the destriped images by the proposed SpcMM is minimized due to the proper estimation of the expected mean and standard deviation for each along-track column. Moreover, SpcMM can automatically destripe an entire Hyperion image cube without the manual selection of defective bands or across-track spatial regions, or the individual selection of band-specific window sizes for spatial smoothing. Two Hyperion datasets acquired over sites in Seal Harbour, Canada, and Coleambally, Australia, have been used to assess the proposed new destriping technique. Examination of the images destriped by SpcMM and SpaMM and the reflectance spectra retrieved from the destriped image cubes reveals that the presented algorithm removes various types of stripes without degrading the images and is superior to the spatial moment matching technique.

Recent Developments in the Hyperspectral Environment and Resource Observer (HERO) Mission

In 1997, the Canadian Space Agency (CSA) and Canadian industry began developing enabling technologies for hyperspectral satellites. Since then, the CSA has conducted mission and payload concept studies in preparation for launch of the first Canadian hyperspectral earth observation satellite. This Canadian hyperspectral remote sensing project is now named the Hyperspectral Environment and Resource Observer (HERO) Mission. In 2005, the Preliminary System Requirement Review (PSRR) and the Phase A (Preliminary Mission Definition) were concluded. Recent developments regarding the payload include an extensive comparison of potential optical designs. The payload uses separate grating spectrometers for the visible near-infrared and short-wave infrared portions of the spectrum. The instrument covers a swath of >30 km, has a ground sampling distance of 30 m, a spectral range of 400-2500 nm, and a spectral sampling interval of 10 nm. Smile and keystone are minimized. Recent developments regarding the mission include requirements simplification, data compression studies, and hyperspectral data simulation capability. In addition, a Prototype Data Processing Chain (PDPC) has been defined for 3 key hyperspectral applications. These are: geological mapping in the arctic environment, dominant species identification for forestry, and leaf area index for estimating foliage cover as well as forecasting crop growth and yield in agriculture.

Toward scene-based retrieval of spectral response functions for hyperspectral imagers using Fraunhofer features

Initial steps are proposed and tested in the development of a method for retrieving and (or) refining instrument spectral characteristics for dispersive hyperspectral imagers such as the Airborne Visible / Infrared Imaging Spectrometer (AVIRIS), Compact Airborne Spectrographic Imager (casi), HyMap, Hyperion, and compact high-resolution imaging spectrometer (CHRIS) based on data acquired by the instrument in operation using statistical spectrum matching with moderate-resolution transmittance code (MODTRAN) modelled instrument results in the vicinity of reference Fraunhofer feature windows. Until now, such scene-based retrieval has focused primarily on refining spectral band-centre shifts while assuming that spectral response function (SRF) parameters remain static. In particular, most methods assume that the SRF is of a Gaussian shape. As a consequence of recent investigations showing that scene-based discernment of SRF shape should be feasible given current typical instrument performance, this paper explores algorithmic components deemed necessary for the development of a look-up table (LUT) based retrieval method for obtaining SRF parameters on a band-by-band basis, even in the presence of minor band-centre or bandwidth deviations from nominal instrument specifications. The proposed method employing these components is appropriate for dispersive hyperspectral imagers but not for others, for example Fourier transform hyperspectral imagers. In experiments using nominal implementations of the proposed components, reference spectra match expected LUT spectra in nearly all cases, even when band-centre and bandwidth deviations are considered. This holds true for all three modelled instruments and nearly all of the six selected Fraunhofer windows. Expected signal-to-noise requirements are in many cases challenging, yet feasible using signal-enhancement techniques such as along-track averaging.

Hyperspectral imagery for mineral exploration: comparison of data from two airborne sensors

Hyperspectral image data sets acquired near Cuprite, Nevada in 1995 with the SWIR full spectrum imager (SFSI) and in 1996 with the Airborne Visible/IR Imaging Spectrometer (AVIRIS) are analyzed with a spectral unmixing procedure and the result compared. The SFSI image has pixels on 1 m by 1.5 m centers, the AVIRIS on 17 m centers; the region imaged by SFSI is a small portion of the full AVIRIS scene. Both have nominal spectral band center spacings of about 10 nm. The image data, converted to radiance units, are atmospherically corrected and converted to surface reflectance. Spectral end members are extracted automatically from the two data sets; those representing mineral species common to both are compared to each other and to reference spectra obtained with a portable IR mineral analyzer. The full sets of end members are used in a constrained linear unmixing of the respective hyperspectral image cubes. The resulting unmixing fraction images derived from the AVIRIS and the SFSI data sets for the minerals alunite, buddingtonite, and kaolinite exhibit strong similarities.

Short-wave infrared (SWIR) imaging spectrometer for remote sensing

The SWIR full spectrum imager (SFSI) is an imaging spectrometer, covering the short-wave infrared (SWIR) from 1200 to 2400 nm, which has been developed for remote sensing from an airborne platform. The sensor has been designed to acquire the full spectrum at high spectral resolution (10 nm) and the full image swath at high spatial resolution (50 cm) simultaneously. The instrument utilizes a platinum silicide (PtSi) detector array, refractive optics, and a transmission grating. A VME bus computer communicates with the array controller, performs the data acquisition, and provides the operator interface. The camera and data acquisition subsystems have been completed and test flown. The fore-optics, spectrograph, and sensor housing have been fabricated. Integration of the camera, spectrograph, and auxiliary components is scheduled for July 1994 followed by laboratory testing and calibration. Our goal is to obtain pilot project data by the end of autumn 1994. Here we describe the optical design, the sensor system, early test flight image data, and expected sensor performance based on laboratory testing. The objectives and procedures for the spectral, geometric, and radiometric calibration of this sensor are also discussed.

Hyperspectral Environment and Resource Observer (HERO) mission

The Canadian Space Agency (CSA) has conducted mission and payload concept studies in preparation for launch of the first Canadian hyperspectral Earth observation satellite. Named the Hyperspectral Environment and Resource Observer (HERO) mission, its objective is to provide information-rich optical imagery that enhances decision-making and stewardship of sensitive ecosystems and natural resources. The mission is designed to provide accurate forest inventory and health information, map the geology of the north, assess environmental impacts, and strategically extend the Canadian investment in Earth observations. The mission builds on the Canadian industry experience and expertise in satellite development and remote sensing and will make new capabilities available for a wide variety of users worldwide. In 2005, the preliminary system requirement review (PSRR) and phase A (preliminary mission definition) were concluded. The resulting mission characteristics are a swath width greater than 30 km, a ground sampling distance of 30 m, a spectral range from 400 to 2500 nm, and a spectral sampling interval of 10 nm. HERO is primarily a flexible tasking mission with a raw capacity of ~600 000 km2 daily ground area coverage. Large-area mapping is to be performed as a background mission. The proposed instrument design consists of dual spectrometers and telescope assemblies. The fore-optics is composed of a three-mirrors anastigmatic (TMA) telescope. The Offner-type spectrometers have separate visible near infrared (VNIR) and short-wave infrared (SWIR) detectors. Expected performance includes a signal-to-noise ratio (SNR) of 600:1 in the VNIR and 200:1 in the SWIR, F/2.2 spectrometers with minimized smile and keystone, and instrument modulation transfer function (MTF) of at least 0.3 at the Nyquist frequency for all wavelengths and fields.

Impact of Sensor Signal-to-Noise Ratio and Spectral Characteristics on Hyperspectral Geoscience Products

Trade-off studies are essential in designing and developing any remote sensing instrument. In order to make effective decisions, the sensitivity analysis of task-specific information to be extracted from the remotely sensed data to the sensors characteristic parameters must be conducted based on various simulated data. The impact of the changes of signal-to-noise ratio (SNR), spectral sampling interval (SSI), wavelength center position (WCP), and wavelength center error (WCE), on the identification and unmixing fraction maps of fifteen selected pure minerals is investigated in this paper. The results show that, for a sensor with a typical responsivity spectral dependence, a specification of SNR = 200 at 2100 nm is insufficient for identifying and mapping the selected minerals with linear spectral unmixing using the SWIR II region (1950 to 2450 nm). On the other hand, if the SNR is high enough, a SSI of 8.2 nm is sufficient for producing these geoscience products no matter where the WCPs of a sensors bands are located. It was found that some minerals, such as dickite, become unidentifiable when the SSI is increased to > 12.3 nm. A WCE has impacts on both the identification and unmixing fractions of some minerals, but these are small relative to the impacts of SNR.

Atmospheric effects on the classification of surface minerals in an arid region using Short-Wave Infrared (SWIR) hyperspectral imagery and a spectral unmixing technique

This study focuses on the comparison of spectral unmixing results from at-sensor radiance data and atmospherically corrected data, i.e., surface reflectance. The airborne Short-Wave Infrared (SWIR) Full Spectrum Imager (SFSI) and Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) were used to acquire data over Cuprite, Nevada, in June 1995 and 1996, respectively. This is an arid region that is well known for its exposed bedrock and alteration zones. For this project, a portion of the SWIR-2 spectral range covering the atmospheric window between 2050 and 2350 nm and a linear spectral unmixing technique were used to map surface minerals, including products of hydrothermal alteration. In total, eight end-members were extracted from the imagery using a new automatic procedure called Iterative Error Analysis (IEA). The atmospheric corrections were applied using a look-up-table procedure implemented in the Imaging Spectrometer Data Analysis System (ISDAS) and created with the radiative transfer model MODTRAN3. The fraction, or abundance, maps derived from the two types of data were compared using the coefficient of determination (R2) and the Average Euclidean Distance Coefficient (AEDC). Very good unmixing agreement was found between the results from the at-sensor radiance data and those from the surface reflectance data. For the SFSI data, the R2 values range from 0.72 to 0.95 and the AEDC values range from 0.098 to 0.023, whereas for the AVIRIS data, the R2 values range from 0.92 to 0.99 and the AEDC values range from 0.021 to 0.008. This suggests that, for the spectral range considered, atmospheric corrections are not necessary for mineral mapping in an arid region and for desert atmospheric conditions.