In the previous topics, we discussed imaging technology and remote sensing systems. Once the data reach the ground, the next question is how to extract information from the images. We’ll begin with traditional photo-interpretation techniques, then move on to the manipulation of digital data.
There are two techniques involved in image processing for remote sensing: enhancing images for presentation and extracting information. Most work at the pixel level and many make use of scene statistics. If the data inhabit more than one spectral dimension (that is, if they have color), then a broad range of techniques can exploit their spectral character and extract information.
Traditional image analysis makes use of certain key elements of recognition, ten of which are developed here. The first four—size, shape, height, and shadow—are related to the geometry of objects in the scene.
Shape
Shape is one of the most useful elements of recognition. One classic shape-identified structure is the Pentagon, shown earlier in Fig. 3.7. In Fig. 3.6, shape alone can identify the airfield, though it is not well resolved. Other instantly recognizable shapes include oval racetracks and the pyramids of Egypt.
Size
Relative size is helpful in identifying objects, and mensuration, the absolute measure of size, is extremely useful in extracting information from imagery.For example, the Hen House radar sites (Fig. 1.4) display characteristic shapes and sizes.
Shadow
Shadows separate targets from background. They can also be used to measure height. See, for example, the Washington monument in Fig. 3.7(b).
Height (depth)
Height is generally derived from shadows in nadir-viewing imagery, but can be derived directly from more oblique views. Stereo imagery has been used to distinguish height in aerial photography and the original Corona imagery. Modern alternatives include LIDAR and interferometric synthetic-aperture radar (IFSAR).
Tone or color
Tone and color are the product of the target albedo and illumination.Consider the difference in tone between the runway (asphalt?) and adjacent regions in Fig. 1.16. The green regions on the south end of Coronado Island (Fig. 1.17) are distinguished from regions of similar brightness by color. On a larger scale, the distinction between dirt and vegetation in San Diego County can also be made by color (Fig. 1.13).
Figure 6.1 SA-2 SAM Base, KH-4 CORONA Product.
Texture
Texture is concerned with the spatial arrangement of tonal boundaries. Texture is the spatial arrangement of objects that are too small to be discerned. Texture depends upon the image scale, but can be used to distinguish objects that may not otherwise be resolved. The relative coarseness or smoothness of a surface becomes a particularly important visual clue with radar data (Fig. 1.23). Agricultural and forestry applications are appropriate for this tool—individual trees may be poorly resolved, but clusters of trees will have characteristic textures.
Pattern
Related to shape and texture is pattern, the overall spatial form of related features. Figure 6.1 shows a Russian SAM site, with characteristic patterns that help detect missile sites, such as the Russian propensity for erecting three concentric fences around important installations. In imagery from systems like Landsat (30-m resolution), irrigated fields form characteristic circular patterns in the American southwest (Fig. 6.2). Geological structures, too, reveal themselves in characteristic patterns, a concept applied in the search for water on Mars and for characteristic textures3 and patterns4 associated with mineral hydra-tion and water flow.
Figure 6.2 Boulder, Colorado. Landsat TM image, bands 4, 3, and 2. The Bighorn Basin is located about 100 miles east of Yellowstone National Park in northern Wyoming. The circle is characteristic of irrigated crops. Bright red indicates the area is highly reflective in the near infrared (TM band 4), which indicates vegetation.
Association
Three elements of photo-interpretation are related to context, or the relationship between objects in the scene to each other and to their environment. These elements are site, association, and time.
Association is the spatial relationship of objects and phenomena, particularly the relationship between scene elements. "Certain objects are genetically linked to other objects, so that identifying one tends to indicate or confirm the other. Association is one of the most helpful clues for identifying cultural features."5 Thermal-power plants will be associated with large fuel tanks or fuel lines. Nuclear-power plants tend to be near a source of cooling water (though this can also be considered an example of site or location). A classic instance of cultural association from the Cold War was the detection of Cuban forces in Angola by the presence of baseball fields in the African countryside (1975-76).
Site
Site is the relationship between an object and its geographic location or terrain. This can be used to identify targets and their use. A poorly resolved structure on the top of a hill might, for example, be a communications relay, based on its location.
Time
The temporal relationships between objects can also provide information, through time-sequential observations. Crops, for example, show characteristic temporal evolutions that uniquely define the harvest. Change detection in general is one of the most important tasks in remote sensing, and follows from this interpretation key.
