Preliminary estimation of photo-interpretation properties of the TerraSAR-X imagery
The TerraSAR-X data are considered as a material for mapping in medium scales (1: 25 000 - 1: 50 000) and the objective of the present work was to estimate the photo-interpretation features of the TerraSAR-X images. It is well known that the photo-interpretation of radar images is substantially different from the photo-interpretation of optical images due to some specific properties of SAR data:
- SAR images are made from a sharp angle
- Speckle noise presence
- Radar shadows on images
- Specific image projection (slant range - azimuth)
- Scattering mechanism of radar sensing differs from scattering mechanism in visible range of wavelength
TerraSAR-X image covered south-western part of Moscow with spatial resolution of 1,5 meter. Signal polarization is VV. Data were acquired on December, 28 2007. Incidence angle is about 39 degrees.
TerraSAR-X image over Saratov with spatial resolution of 1 meter. Signal polarization is VV. Data were acquired on December, 24 2007. Incidence angle is about 44 degrees.
TerraSAR-X image over New Urengoy with spatial resolution of 1 meter. Signal polarization is VV. Data were acquired on December, 27 2007. Incidence angle is about 44 degrees.
The estimation of photo-interpretation features was carried out using three images:
All the TerraSAR-X images were acquired in winter.
The photo-interpretation of the images was done using PHOTOMOD system.
The images were loaded as 8-bit data in GeoTiff format and then the brightness and contrast were adjusted.
The photo-interpretation was done by several operators who had not had any experience with SAR data processing before, though they had been interpreting optical images for a long time.
The best visual perception for non-corrected images was achieved when viewing with resolution 2-3 pixels to resolution sell on the PC monitor. In this case image on monitor equals approximately to a map in 1:12 000 -1: 18 000 scale.
The photo-interpretation was carried out without any filters.
|Fig 1. TerraSAR-X image over Moscow area with 2-meter spatial resolution
||Fig 2. Optical image over Moscow area with 1-meter spatial resolution
Fig 3. Buildings at the TerraSAR-X image (left) and optical (right)
Summarized results of the photo-interpretation are presented in tab. 1
Brief characteristics of objects
Buildings with the area of up to 1500 m2 in urbanized zones are almost never recognized. Buildings with the area of 1500-2000 m2 can be recognized and vectorized with the 90% probability. It is important to note that there is no clear boundary between the wall and the roof, unlike for the optical imaging. This is the reason why the positional location accuracy
is not high enough. Boundaries of residential areas can be recognized with adequate accuracy.
Fig 4. Different type of roads at the 1-meter TerraSAR-X data (left) and 1-meter optical imagery (right)
Ten meters wide roads can be recognized with the 70% probability.
20-30 meters roadways usually can be recognized and vectorized in 90% cases. Passages between boulevards are clearly visible althougha boundary between carriageway and grass-plot can't be always seen and looks like a solid roadway. The possibility to detect passages depends on
their orientation, for example, better results can be achieved if a passage locates in range direction.
Fig 5. Road junction at the 1-meter TerraSAR-X (left) and 1-meter optical imagery
Multilevel road junctions are easily recognized, which is due to height difference between traffic ways, but there is a difficulty to determine a number of branch lines to a road junction.
Fig 6. Railways at the 1-meter TerraSAR-X (left) and 1-meter optical imagery
They are recognized easily enough. Difficulties can be caused by radar shadows for the images where a railway crosses urban areas. The number of railway tracks can not be recognized.
Fig 7. Hydrography objects at the 1-meter TerraSAR-X (left) and 1-meter optical imagery
Rivers with the width of up to 10 meters can be detected in 2/3 of cases. Their visibility strongly depends on the location regarding the line-of-sight direction and objects on the sides that can hide a river: buildings, trees etc. 20-30-meter rivers have been recognized easily enough on analyzed images.
Water objects with the area of about 200 sq. km are also detectable, but those covered by snow are recognized worst. Also ships on water surface can be seen clearly.
Fig 8. Bridges at the 1-meter TerraSAR-X (left) and 1-meter optical imagery
Bridges are clearly visible over the water area, but over land their recognition is more difficult. The presence of radar shadow makes the location accuracy worse.
Fig 9. Vegetation at the 1-meter TerraSAR-X (left) and 1-meter optical imagery
Vegetated areas can be recognized and vectorized easily enough. An experienced expert can detect and distinguish wood and meadow vegetation. It is important to take into account the winter season when the radar data have been acquired.
We have compared our results with QuickBird data take from Google Earth, and also with topographic maps in 1: 25 000 scale. The comparison revealed some mistakes in our data interpretation. For example, open pool was recognized as a building, asphalted areas in parks were interpreted as buildings as well. Ponds were mistaken for vegetation.
Fig 10. Recognized 2-meter TerraSAR-X image over Moscow area
We have divided into three parts the objects by degree of its recognition:
| Easily recognized objects
|| Buildings with the area of 1500sq. m, roads of 20 meters wide and wider, multilevel road junctions, railroads, rivers of 10 meters wide and wider, vegetation areas of 10 000 sq. m and more
| Moderately recognized objects
||Buildings of 1000 sq. m, roads up to 10 meters, rivers of 10 meters, vegetation with area in range of 5000-1000 sq m
|Poorly recognized objects
||Buildings with area up to 1000 sq m, roads up to 10 m, rivers up to 10 meters, hydrography areal objects up to 1000 sq. m, pipelines
We managed to recognize the most part of the objects which are depicted on maps in 1: 25 000 - 1: 50 000 scales. It is necessary to have an additional material (maps etc.) for some objects.
In order to increase the quality of the radar images recognition it is recommended to follow the next guidelines:
- Try to remove the speckle structure from images using different filters, and to investigate the possibility of different polarization usage.
- Correct orientation of the image makes easier the photo-interpretation and can increase its reliability
- It is necessary to train operators in SAR data processing, photo-interpretation and vectorization
A well-organized library of examples of radar image data interpretation could be of great help.