| On the Internet: SAR Feature Extraction
By Mark R. Lucas Overview
This article shows how SAR imagery can be prepared to enhance features in the scene and be appropriately mapped so it is pleasing to the human eye The imagery used in this project was made available online at the Jet Propulsion Lab (JPL) Radar Imaging home page as a part of a contest that was run on the Internet. Using standard applications on a Macintosh Power PC, I was able to prepare the imagery and was eventually selected as the winner of the contest. The Problem
SAR has enormous potential due to its all weather, day and night capabilities. Since the spatial resolution is derived from the frequency of the transmitted signal, the detected imagery has very high relative accuracy. Despite this, one often hears complaints that SAR is difficult to interpret - that it is not as literal as the passive sensors.
Properly prepared, SAR can be visually interesting and provide information that is not readily available with passive sensor platforms. This example explores some interesting areas of feature extraction using the active SAR signal return as a false elevation map. How it was Accomplished
The SIR-C mission recently collected the first multi-band multi-polarity synthetic aperture radar data and much of this data is displayed on the JPL Radar Imaging home page which can be found at http://southport.jpl.nasa.gov/. Bruce Chapman maintains the site and began a contest by placing seven bands online and posting the rules. They were simple: 1. You must use one or more of the input bands; and, 2. You must describe what you did. Seven bands were supplied using the following frequencies: L band - 21 cm; C band - 5.6 cm; and X band - 3 cm. They consisted of the following polarities: hh - horizontally transmitted, horizontally received; hv - horizontally transmitted, vertically received; and vv - vertically transmitted, vertically received. Creating the Image
My strategy was to use all seven bands of the SIR to create an interesting image of Sunbury, Pa. The trick to doing this was to map the bands into numerous perceptual channels. I decided to use L_hv, C_hv, and X_vv for the red, green and blue channels, respectively. My selection was simply on the basis of attempting to map the wavelengths from longest to shortest into the optical band; red longest, blue shortest.
With the rest of the channels, I wanted to generate an elevation map. This map probably has little correlation to the actual elevation of Sunbury, but could serve as a cue to the relative strength of the radar returns across those bands. Areas where there are strong returns are elevated (the city of Sunbury) and areas where there is little return are at low spots (the river for example). Steps Used in Preparing the Data
I began by opening Adobe Photoshop v 3.0 and loading the L band hv into the red channel, the C band hv into the green channel and the X band vv into the blue channel.
Three bands were used to create the RGB texture map shown below using Adobe Photoshop.
The remaining bands were then combined using the multi-channel mode of Adobe Photoshop. Since I intended to sample this image at every five pixels to derive elevation, I ran a Gaussian Blur with a five pixel radius to avoid sampling errors.
Infini DTM is a powerful 3D modelling and visualization program from Specular Inc. One of the features that I like to use is the terrain object. The terrain object is a wire mesh that can be texture mapped and elevation mapped with imported pictures. Using the texture and elevation maps shown above, I created a 3D model and rendered the following images. Sunbury, Pa.
Scientists are using this radar image of the area surrounding Sunbury, Pa. to study the geologic structure and land use patterns in the Appalachian Valley and Ridge province. This image was collected on Oct. 6, 1994 by the Spaceborne Imaging Radar-C/X-Band Synthetic Aperture Radar (SIR-C/X-SAR) on orbit 102 of the Space Shuttle Endeavor. The image is centered on latitude 40.85 degrees north latitude and 76.79 degrees west longitude. The area shown is approximately 30.5 km by 38 km (19 miles by 24 miles). North is toward the upper right of the image. The valley and ridge province occurs in the north-central Appalachians, primarily in Pennsylvania, Maryland and Virginia. It is an area of adjacent valleys and ridges that formed when the Appalachian mountain range was created some 370 to 390 million years ago. During the continental collision that formed the Appalachians, the rocks in this area were pushed from the side and buckled much like a rug when pushed from one end. Subsequent erosion has produced the landscape we see in this image. The more resistant rocks, such as sandstone, form the tops of the ridges which appear as forested greenish areas. The less resistant rocks, such as limestone, form the lower valleys which are cleared land and farm fields and appear brown. Smaller rivers and streams in the area flow along the valleys and in places cut across the ridges in "water gaps." In addition to defining the geography of this region, the valley and ridge province also provides this area with natural resources. The valleys provide fertile farmland and the folded mountains form natural traps for oil and gas accumulation; coal deposits are also found in the mountains. The colors in the image are assigned to different frequencies and polarizations of the SIR-C radar as follows: red is L-band horizontally transmitted, vertically received; green is L-band horizontally transmitted, vertically received; blue is X-band vertically transmitted, vertically received. The river junction near the top of the image is where the West Branch River flows into the Susquehanna River, which then flows to the south-southwest past the state capitol of Harrisburg, 70 km (43 miles) to the south and not visible in this image. The town of Sunbury is shown along the Susquehanna on the east just to the southeast of the junction with West Branch. Three structures cross the Susquehanna; the northern and southern of these structures are bridges, the middle structure is the Shamokin Dam which confines the Susquehanna just south of the junction with West Branch. Summary
SAR has enormous potential in numerous application areas. Similar to color composite TM bands, multi-band SAR can be mapped into the primary colors to produce visually pleasing image sets for analysis. Additionally, the active nature of the sensor opens up new areas for feature extraction by mapping signal returns into false elevation models. Acknowledgements
The first implementation using detected SAR as a false elevation map originated in the Digital Processing Labs of Harris Corp. in Melbourne, Fla. Dave Bell, Ph.D., and Todd Jamison (currently with Observera Inc.) came up with the original concept which was subsequently refined by Ken Debes. Infifi DTM is a product of Specular Corp. and Adobe PhotoshopTM is produced by Adobe Corp. The PowerPC is manufactured by Apple Corp., and the analysis on Sunbury was obtained from a JPL press release. About the Author:
Mark Lucas is the chief technical officer of ImageLinks Inc., based in Melbourne, Fla. He recently retired from the U.S. Air Force where he worked in various aspects of classified remote sensing including launch and ground processing, cross sensor product generation and product distribution. ImageLinks provides high volume automated value added processing for the commercial remote sensing market. He may be reached at 407-722-4550 (phone); 407-722-4588 (fax); or E-mail: [email protected] Back |
