At Applied Analysis Inc., we have a long history of developing software tools to exploit remotely sensed geographic imagery. The current version of our Subpixel Classifier spectral image processing tool allows users to develop spectral signatures from multispectral and hyperspectral imagery and apply them to locate materials of interest within a scene, as well as scene-to-scene. The unique subpixel detection capability of this software allows the user to detect materials that occupy as little as 20% of a pixel and correctly distinguish spectrally similar targets, such as a particular vegetative species, interspersed with other vegetation. The software1 has been used in a wide number of applications, including urban land cover studies2, wetland tree species identification3, crop identification4, environmental monitoring5, wetlands identification6, and pest infestation location7.
A U.S. Government customer contracted with Applied Analysis Inc. to use our subpixel classification technology to create land use analysis products from commercial satellite imagery. The customer’s area of interest covers millions of square miles, and dozens of Landsat Thematic Mapper image scenes were required to be processed. Additionally, the customer was interested in how land usage changed over time which dictated that archival satellite imagery be processed to develop a historical record.
The Subpixel Classifier image exploitation tool produces a classification image from an input multispectral image. Each pixel within the classification image is assigned a class value which is representative of the target material pixel fraction. Signatures were developed for very specific land cover features and employed operationally. The resulting classification images are typically somewhat sparse since pixels within the image are set only when the target material is present at that location on the ground. The image analyst overlays these classification images on the original image so that they can correlate material locations with geographic features and locations. Both the original image and the classification image can then be georectified and georeferenced to form a land use map of the area.
This information can be combined with other geographic information and stored in a GIS. A given location would then have imagery and image-based products from multiple dates coupled with other geographic features such as political boundaries and characteristics, rivers, roads, etc. The resulting GIS is a powerful tool for exploring the area of interest, visualizing spatial and temporal patterns, and planning operations. Recognizing this power, the customer wanted to be able to disseminate this information to interested parties scattered throughout the world. Many of the end users are not image analysts or GIS specialists. They have neither the tools nor the expertise to analyze an image-based GIS, yet they have a need for customized maps based on specific products of interest.
The solution to this need was to use Esri’s ArcIMS software to serve up custom maps based on image exploitation products and a GIS data base. The end user need only a standard web browser and network connectivity to be able to download imagery and processed data products. The web interface allows straightforward access without requiring special hardware and software, or specialized technical knowledge. The end result is a Geodata Visualization System (GVS).
From the user’s standpoint, GVS is a set of dynamic web pages that allows them to focus in on an area of interest, select the products (overlays) they are interested in, and create a printed output. They typically begin with the GVS Products web page. This page lists available products by country. Hyperlinks within this page take you to the GVS web page containing products for that area. They can also jump to the GVS Global Entry page and browse for product areas graphically. Links to pages containing more information about GVS and the sponsoring project are also provided on the Products web page.
The remainder of this paper describes the objective and requirements
for GVS and the resulting design features of the system. Some comments
on system performance are also provided.
The GVS interface must allow straightforward access without requiring special hardware and software or specialized technical knowledge. End users are primarily interested in information in map form. This form aids their visualization of geography and assists in operations which require the user to visit locations of interest.
The basic requirements to use GVS are:
Many, if not all, organizations connected to SIPRNET do not allow
mobile code such as Java applets to move through their firewalls. GVS does
not use Java or the ArcIMS Java client viewer. Only dynamic HTML and JavaScript
are allowed in GVS.
Each area site uses its own ArcIMS map service and web site directory. Consequently there are a number of site directories that would normally contain duplicate HTML code, JavaScript code, and graphical images. To aid in the maintenance of the overall web site, the code supplied with ArcIMS was modified so that most of the HTML and all of the JavaScript and graphical images could be placed in a common code directory. Each area site contains an area-specific JavaScript file containing area specific parameters, its own default.htm file, viewer.htm file and MapFrame.htm file. Otherwise the code is stored in a common directory accessible by each site. This makes it easy to update the code when changes are needed.
All area sites contain at least one base image and many sites contain several images. The original image data set contained layers acquired at multiple wavelengths. This spectral information is the key to deriving additional information from the data using spectral signatures. Prior to including these images on the GVS site, the images are converted to a true color representation and stored as PNG files. Each image also has an associated world file. The PNG image format was used since it support transparency.
Many imaging systems, such as ERDAS IMAGINE, use a multi-resolution pyramid layer scheme to speed up image zooming. When the user requests a low resolution, wide area version of the image, the system accesses the pre-calculated low resolution version within the input file. As the user zooms in, the system accesses successively higher resolution layers, thereby optimizing access speed. It was hoped that we could take advantage of the multi-resolution pyramid layer file that comes with ERDAS IMAGINE; however, ArcIMS does not seem to utilize that feature. Future work will attempt to improve image access speed by using LizardTech’s MrSID encoding software which also employs a multi-resolution image encoding scheme.
New data are added to the system infrequently so it is acceptable to
store content in static flat files such as image files and shape files.
Classification results are stored in image format to save disk space. The
down side of this is that the user cannot query the material pixel fraction
for a given location. This was considered an advanced feature that may
be added in the future. As the number of area sites increases in the future,
it may be necessary to implement a data base access solution using ArcSDE.
This would reduce the number of map services and site directories required,
making the system more maintainable.
Figure 1. Layout of the GVS Web Site
The GVS global navigation site is a wide-area map with highlighted areas to indicate where products are located geographically. If the user selects one of these highlighted areas, and they will be taken to the GVS web page for that area. From there they can navigate around, query features, and produce a customized printed output.
The GVS global navigation page is divided into functional areas as shown in Figure 2.
Figure 2. GVS Global Navigation Page
Navigation tools, which are located along the left side, allow the user to adjust the extent and appearance of the map. The regular ArcIMS insert map can be toggled on and off. A unique feature of GVS maps is that the user can toggle a latitude/longitude grid on and off. The typical pan and zoom tools are also provided. In addition, the user can zoom by fixed increments using the buttons along the left side of the map area.
Figure 3. Navigation Tools
Data tools are arranged along the top of the map area and allow the user to create a printable map and get usage help. The user cannot print the displayed map directly. They must create a printable map using Print Map feature. This tool creates a new web page with the current view and overlays. When the user selects Print Map, a secondary dialog appears as shown:
Figure 4. Print Dialog
The user has the option of producing a Regular Print Page or selecting Advanced Options. The Advanced Options feature allows the user to render the printable page at higher resolutions and to save the rendered map image to a file, as shown in Figure 5.
The user can change the current Map Mode from navigation to selection
by selecting one of the Product Selection Tabs located at the top of the
map area.
Two categories of products are available:
Figure 6. Sample Area of Interest Site Layout
Navigation tools are expanded slightly, as shown in Figure 7, to give the user more control over the map settings. The Slider button is an optional tool and does not appear on all Area of Interest sites. If it is present in the Navigation toolbar area, the uppermost overlays can be slid back to reveal the bottom overlay by pressing and dragging the left mouse button within the Map Area. By rapidly dragging the mouse, the user can visually correlate features in the upper overlays to those in the bottom overlay. This is useful for visual change detection within two overlaid images or for associating features in an image with features in an underlying map. Note that the slider visibility window does not appear on the printable map page. This feature is only available for interactive use.
Figure 7. Area of Interest Navigation Tools
The Data Tools section is expanded to allow the user to inquire about features. Every tool button has a help portion indicated by a question mark. If the user selects that portion of the button, a small help window appears with specific instructions for that button.
Figure 8. Area of Interest Data Tools
The Measure tool allows the user to measure distances on the map and display segments. The Select tool allows the user to select features by dragging the mouse within the map area to define a rectangle. Any feature within the Active Overlay will be selected and information about those features will be displayed in a separate pop-up window. The Identify tool is similar in that it allows the user to select a single feature within the Active Overlay.
Overlays are layers of image data stacked on top of each other. Overlays are generally semi-transparent. In those areas where there is no content you can see down to the underlying overlays. Areas with content block the underlying overlays from showing through. The user can control overlay visibility and the order of the overlays in the stack.
A unique feature of GVS is the ability to include and exclude overlays from the overlay list located along the right side of the map area. The Overlay Control button brings up a separate window containing information about the overlays. The user can change the order of the overlays and include or exclude overlays from the current set. They can also view additional detailed information about each overlay.
Figure 9. Overlay Control Dialog
The design philosophy was to place minimal requirements on the GVS client (user hardware) by performing most of the image manipulations on the server side. This approach requires a dedicated and capable server to meet the demand. The GVS system is also designed to be inherently scalable. The GVS system hardware consists of a Compaq Proliant DL380 Server with dual 866-MHz Pentium III processors, 1 GB RAM, an onboard RAID controller, and dual (redundant) power supplies. RAID level 5 is used in the disk array to ensure speed and data integrity, as well as flexibility to provide future expansion. The hardware is rack-mounted.
The GVS server runs the Microsoft Windows NT Server, Version 4.0, operating system with Service Pack 5 installed. The HTTP server software is Netscape iPlanet Web Server, Enterprise Edition, Version 4.0. The iPlanet web server software is a multi-process, multi-threaded secure web server providing high performance, reliability, scalability, and manageability. The server supports remote administration, access control, Secure Sockets Layer (SSL) [not currently used], server-side Java servlets and JavaScript, and flexible logging and performance monitoring. Java servlet support is native to iPlanet and a separate servlet engine installation was not required.
GVS’s special map-based image content is assembled and supplied to the iPlanet web server using Esri’s ArcIMS software. ArcIMS is a suite of software modules that involve the client, the server, and data management. On the client side, Dynamic HTML and JavaScript are used. Client-side Java applets are not required. On the server side, ArcIMS provides an Application Server module and special Java servlets. A specific configuration of these servlets forms a Map Service. In the GVS system, the communication between the Web Server and ArcIMS Application Server takes the form of Servlet Connectors which are ArcIMS XML exchanges. This architecture was laid out in Figure 10. ArcIMS uses the native servlet engine in iPlanet and adds spatial servers and the application server.
Figure 10. Schematic of ArcIMS Distributed Processing System
A typical Landsat image contains approximately 6000x5000 pixels. When converted to true color representation and compressed in a PNG file such an image only occupies about 20-30 MB of disk space. Such images normally display within about 8 seconds, not counting download time. The typical downloaded image is about 100-150 KB, depending on the client’s screen resolution and the image content. This download occurs each time the user adjusts the map extents or modifies a layer. For users connecting via a modem, this time represents the major delay experienced. For users connecting via a high-speed connection, the major delay is the time required for the ArcIMS image server to load the image and create an output map.
The largest area site fielded to date consisted of 5 overlapping Landsat scenes mosaiced together to form a single image approximately 14000x9000 pixels. The pixel size was interpolated to twice the normal size to reduce the number of pixels. The resulting compressed file size was approximately 50 MB. This site can take up to 30 seconds to display, not counting download time. This time delay is considered unacceptable for most users.
Long display time is currently the major system issue with GVS. The time required for the ArcIMS imageserver to load image data from file and produce an output map image is marginally acceptable in most cases and unacceptable for very large images. It is anticipated that version 3.1 of ArcIMS will have improvements in imageserver speed.
We have also done some work with LizardTech’s MrSID software which employs a multiresolution image storage format in the hopes that it would speed map display. So far, given limited testing and tuning, results are mixed. Initial map display is somewhat slower, but map refreshes are somewhat faster.
Following initial deployment of the system it became clear that one maintenance issue would be version control of the customized ArcIMS client software (HTML and JavaScript) for multiple site directories. This issue was alleviated somewhat by modifying the code to use a common code directory. Each area site refers to a single code directory for basic HTML and JavaScript along with the GIF files that are used in the interface look and feel. This has greatly simplified version control and system updates.
2) "Regional Ecosystem Analysis: Puget Sound Metropolitan Area," American
Forests, 7/25/98.
Available at http://www.discover-aai.com/whitepapers.htm.
"Regional Ecosystem Analysis: Chesapeake Bay Region and Baltimore-Washington
Corridor,"
American Forests, 3/10/99. Available at http://www.discover-aai.com/whitepapers.htm
3) R. Huguenin, M. Karaska, D. Van Blaricom, and J. Jensen; Photogrammetric Engineering & Remote Sensing Vol. 63, pp. 717-725, June 1997.
4) http://www.discover-aai.com/cropdetect.htm
5) http://www.discover-aai.com/fuel.htm
6) http://www.discover-aai.com/waterway.htm and http://www.discover-aai.com/wetland.htm
7) http://www.discover-aai.com/pest.htm and "An Evaluation of the Utility of Subpixel Analysis of Thematic Mapper Imagery for the Spruce Beetle Outbreak on the Manti-LaSal National Forest," J. Johnson, P. Greenfield, and A. Steve Munson, published June 23, 1998.
Dr. Gene Roe Director Of Business Development
Applied Analysis Inc.
630 Boston Road Billerica, MA 01821
Phone: (978)663-6828
Fax: (978)663-6389
groe@discover-aai.com