Visit PANCROMA Website

See additional satellite image processing articles at menu selection 'White Papers'

The author would like to thank DigitalGlobe® for kindly providing the imagery used in this study, and the IEEE GRSS Data Fusion Technical Committee for organizing the 2012 Data Fusion Contest.

WorldView-2® Spectral Analysis

Some of the important features of remotely sensed multispectral data include resolution, the number and spectral sensitivity of the bands, dynamic range, spatial and temporal coverage, degree of data processing, and other factors. One of the problems with multispectral satellite data available from most of the free data types (Landsat, SPOT, EO-1) is resolution. Among those mentioned, SPOT has the highest resolution at 20m. However, SPOT has a very limited band count of only four (green, red, NIR and SWIR). Low resolution and/or limited coverage of the spectral range (especially the useful infrared frequencies) can be a significant drawback for some applications, for example feature change and mineral prospecting.

Although DigitalGlobe® WorldView-2 data is not free, it offers some attractive features, including 2m resolution for the multispectral bands and 8 individual bands including three NIR bands. Depending on what you are trying to do, this data may be worth an investment. This article outlines how WorldView-2 multispectral data can be processed using PANCROMA™.

WorldView-2 multispectral data is offered in a single file, non-contiguous (Planar Configuration=2) GeoTiff format. This means that all eight multispectral bands are bundled into a single GeoTiff file. Sixteen bits of dynamic range are provided. This format is a little complicated to parse from a programming perspective, especially for spectral signature analysis where all eight data files must be accessed at the same time, rather than in sequence. As a result, the first step for processing using the PANCROMA™ utilities is to decompose the single GeoTiff file into eight separate 16-bit band files.

To do this, open the single bundled GeoTiff file as issued by DigitalGlobe® by selecting 'File | Open'. Then select 'Display One File' | 'Decompose Eight File Digital Globe Bundle' | 'Save as 16-Bit'. (Note that there is an option to save the bundle in 8-bit format. It is advisable and necessary to save in 16-bit format to preserve the full precision of the dynamic range of the digital numbers in the band files.) When prompted, give the band files a root name and a numerical suffix will be attached designating the band numbers when the eight individual GeoTiff band files are saved.

PANCROMA™ Point Spectrum GeneratorTM

The PANCROMA™ Point Spectrum GeneratorTM is a tool that computes a spectral signature based on user inputs by clicking on a few pixels of known physical characteristics (for example a particular vegetation type). Because calibrated surface reflectances are computed, comparisons can be made within a single data bundle or across different data bundles. PANCROMA™ will compute a spectral signature that can be used to help identify all pixels with similar characteristics. To use this tool, select 'File' | 'Open' as usual. Next select 'Spectral Analysis' | 'WorldView-2 Point Spectrum Generator' | 'Eight 16-Bit Bands'. When you do so a band selection screen will be displayed. Note that there are two radio buttons labeled 'Display Single Band' and 'Display Color Composite'. This allows the choice of using either a single multispectral band for point selection or a composite of three bands. For this exercise select 'Display Single Band'. You can select the multispectral band that you want to be displayed for point selection by checking the appropriate radio button.

[Point Selection Screen]

Band Display Selection Form - lets you specify the image type that you will be selecting your target pixels from You can choose either a grayscale band image or a natural or false color RGB image.

When you click 'OK', the selected band image will be displayed, along with the Point Spectrum Data Entry Form, as shown below. Note that for this example I decided to use the band4 image for making my point selection. This is often a convenient choice because band4 is less susceptible to Rayleigh scattering than the shorter wavelength bands, and therefore can offer a clearer view of surface features.

[Point Spectrum Form]

Point Spectrum Form - this is the display for selecting target pixels. You add pixels to the list by clicking with your mouse. All of the pixels on the list will be converted to Top of Atmosphere (TOA) reflectance values and plotted versus band number.

For this simple example, we will generate the spectrum of the highly reflective dock structure at the upper part of the image. We can do this by successively clicking the cursor on the structure in whichever band file you selected for display. Note that you can zoom into the image if necessary in order to identify target pixels. The image below shows the dock structure along with the approximate positions of my selected points (shown as red dots). I used the default band4 image. You can sample as many as 50 points, although I only used seven for this example. Note that the coordinates of your selection appear in the text boxes and a running count is kept of your selections.

[ Grond Targe]

Target Pixels - I have chosen a seven pixel sample as representative of my target. Their locations are indicated by the red dots.

When you have collected your sample, click 'OK'. The TOA reflectance form will appear for band1 as shown below. You must enter the Acquisition Date, Solar Elevation Angle, Absolute Radiometric Calibration Factor and the Effective Bandwidth from the WorldView-2 metadata (IMD) file so that TOA reflectances can be computed.

[TOA Reflectance Data Form]

TOA Reflectance Form - Each WorldView-2 multispectral band typically has different Absolute Calibration and Effective Bandwidth values. These must be obtained from the WorldView-2 metadata and manually entered into the form, one pair for each band.

The computation of TOA reflectance for WorldView-2 data is described in the Digital Globe publication entitled Radiometric Use of WorldView-2 Imagery. This publication is available at the Digital Globe website and on the PANCROMA™ Free Data web page. Additional information regarding IMD files and calibrating WorldView-2 data is shown in Section 34 of the PANCROMA™ Instruction Manual. Such a form will appear for each band since the Absolute Radiometric Calibration Factor and the Effective Bandwidth values will generally differ for each band for WorldView-2 data. (However the Acquisition Date and the Solar Elevation Angles will be the same for each band and need be entered only once.) When you click 'OK' on this form for band8 (the final one presented), the spectrum will be generated.

[Point Spectrum]

Target Spectrum - the TOA reflectances have been computed for each of the selected pixel Digital Numbers. They are plotted versus band number. This spectral signature can be used by the PANCROMA™ Spectral AnalyzerTM to identify all similar pixels in the image.

The plot shows the computed TOA reflectances for each of the selected points versus band number. The average reflectance for each band is shown in red. The maximum and minimum reflectance values as well as the average reflectances for each band are also echoed to the diagnostics screen. Now you can analyze the plot. The spectrum has several unique features that can be exploited for classification of the image. There are two possible ways to proceed in order to classify the target areas: using the PANCROMA™ Spectral AnalyzerTM or measuring the Euclidean Distance. The next section will demonstrate how the Spectral AnalyzerTM can be used to isolate all pixels in the image that match the spectrum of the identified points.

PANCROMA™ Spectral AnalyzerTM

The PANCROMA™ Spectral AnalyzerTM is a tool that can identify candidate pixels in the image that correspond to a target spectrum. This is done by setting selection criteria for each band. The criteria will be based on the spectrum generated in the previous section. We can instruct PANCROMA™ to search for all pixels either greater than or less than the band target for each individual band. The Spectral AnalyzerTM will identify all pixels that conform to the selection criteria.

To do this, select 'File' | 'Open and open the eight WorldView-2 band files, band1, band2, band3, band4, band5, band6, band7 and band8. (WorldView-2 spectral band data is given in Section 34 of the Instruction Manual). Now select 'Spectral Analysis' | 'WorldView-2 Spectral Analyzer' | 'Eight 16-Bit Bands' | 'Manual Method'.

The Spectral Criteria Form will be displayed. The Spectral Criteria form has a track bar for each WorldView-2 band. There are also radio buttons that allow you to set the selection parameters. With the default 'Must Exceed' setting, all band pixels that exceed the value set on the track bar will be selected. If the 'Less Than' radio button is selected, all pixels with DNs less than the track bar setting will be selected. When the 'Include' check box is checked, the band is included in the computation. If not, the band is ignored. PANCROMA™ will Boolean AND the selection criteria and will report all pixels that satisfy the conditions.

Note that PANCROMA™ has automatically entered the spectral signature values determined by the Point Spectrum GeneratorTM. Set your selection criteria using the track bar sliders. In this case, I left all the band radio buttons set on 'Must Exceed'. I then adjusted the track bar sliders so that each indicated value was equal to the minimum reflectance in each group. Note that it might make more sense to compute the standard deviation (sigma) and set the slider at the average minus 2X or 3X sigma but the reflectances were fairly tightly clustered so I simply selected the minima. When you are finished setting your selection criteria, select 'OK'. Afterwards, a data entry box will be displayed for entering the appropriate band1 TOA Reflectance data.

[Spectral Analyzer Form]

Spectral Analyzer Form - lets you input the selection criteria for comparing all the pixels in the band files to the spectral signature. The average TOA reflectance for each band from the previous analysis have been automatically entered into the form.

Because the calibration factors and effective bandwidth values are different for each WorldView2 band, they will typically have to be entered eight times. However, if you are transferring a spectral signature from the Point Spectrum GeneratorTM as in this example, PANCROMA™ will store the metadata and enter it automatically into the TOA Reflectance Data Form. (Note that the solar elevation angle and Julian day data only has to be entered once. The appropriate data entry form will appear eight times to allow input of the values. (The data can be found in the .IMD file that accompanies each WorldView-2 data set.) It is also a good idea to check the 'Retain settings between runs' check box on the Spectral Criteria form. That way, all of your data will be saved so that you will not have to re-enter anything if you are not satisfied with your first run and need a second (or third). When you select OK the final time (for band8), PANCROMA™ will compute the pixels meeting the selection criteria and color them red. It will also compute the percentage of red pixels in the image and display this number in the Main Window dialog box.

The image below shows the results of the analysis. Spectral AnalyzerTM correctly identified the pixels matching the spectrum on the dock structure, and also identified other structures (mostly rooftops) also matching the spectrum. The target pixels are highlighted in red.

[Spectral Analyzer Plot]

Spectral Analyzer Plot - all the pixels in the image that match the selection criteria are highlighted in red.

PANCROMA™ Euclidean Distance Analyzer

An alternative way to match spectral signatures is distance minimization. In this technique, the multispectral band data is considered an n-dimensional vector, with n being the number of available spectral bands. This is the so-called hypercube approach, i.e. three orthogonal vectors defining the edge of a standard cube and additional orthogonal vectors defining a hypercube. In this example there are eight spectral bands so the distance to the target vector will be minimized in 8-space.

The method computes the distance between each multispectral reflectance vector and the target vector defined by spectral signature you are trying to match. The target vector is entered into the Spectral Criteria form automatically, using the Point Spectrum GeneratorTM as described previously.

To conduct a distance match, open the eight WorldView-2 I16 band files again. Then select 'Spectral Analysis' | 'WorldView2 Spectral Analyzer' | 'Eight 16-Bit Bands' | 'Euclidean Distance'. The Spectral Criteria Form will become visible. If you used the Point Spectrum GeneratorTM, the target vector will already be entered in the sliders as the averages of the selected pixels for each band. You can of course modify these if you wish. In this example I had previously adjusted the track bars to match the spectrum minima, so I reset them back to the average values that were initially input. When you click 'OK' the band1 TOA Reflectance Data Form will again become visible. The metadata that you originally entered when you generated the Point Spectrum should still be stored. Click 'OK' for each band screen.

When you acknowledge the final band8 screen, the Euclidean Distances will be computed. They will be displayed as a color contour plot and a separate grayscale image. In the color contour plot, the closest Euclidean Distance (best match) will be in red. For the grayscale image, the closer the Euclidean Distance, the smaller the grayscale value, meaning that a grayscale value of zero (black) represents the closest match. The color contour plot and grayscale image for this analysis is shown below. The dock structure and surfaces with similar reflectance spectra are red, or black in the grayscale image.

[Euclidean Distance Plot]

Euclidean Distance Color Contour Plot - The Euclidean Distances to the target vector (closeness of match statistic) have been computed for each pixel in the data set. The closest matches are red.

[Euclidean Distance Plot]

Euclidean Distance Grayscale Contour Plot - grayscale contour plot. The closest distances are black.

Comparison of the Spectral Analyzer and the Euclidean Distance plots show that they agree on the match for the dock structure. The Euclidean Distance plot is a bit more selective than the Spectral Analyzer. In order to determine which is more correct, additional Point Spectra of target pixels can be run, or more preferably calibration with spectra collected on the ground can be conducted.

This article has presented a few ideas for multispectral analysis of WorldView-2 data. PANCROMA™ multispectral analysis tools plus WorldView-2 high resolution data provides interesting possibilities for many types of prospecting and analysis. Multispectral analysis of satellite data is a powerful technique and PANCROMA™ offers some useful tools to get the most out of this data.

Web site and all contents Copyright TERRAINMAP Earth Imaging LLC 2010, 2011, 2012. All rights reserved.