Spectral Processing

Spectral data processing extracts information from spectral image data. Multispectral and hyperspectral imagery from overhead sensors capture the earth's surface in several spectral regions. By analyzing the reflected light at several wavelengths we can determine landcover types and properties.

Spectral sensors take advantage of transmission windows in the earth's atmosphere that allow sunlight to penetrate and illuminate the surface. Incident electromagnetic energy in the visible, near infrared, and shortwave infrared regions is reflected back to the sensor. The reflectance properties of landcover materials can be used to identify and characterize the materials. Multispectral sensors capture a few relatively broad spectral bands whereas hyperspectral sensors capture hundreds of narrow spectral bands. Hyperspectral sensors provide finer spectral resolution at the cost of increased noise and volume of data to be processed.

The spectral sensor captures not only the reflected energy from the surface, but also energy scattered by the atmosphere. The two sources combine additively to form the sensor response. Spectral processing must correct for the atmospheric and any sensor contributions to determine the surface reflectance and surface brightness (see Image Calibrator). A spectral return is acquired for each pixel and each pixel represents a mixture of material reflectances averaged together. To determine individual material contributions, the pixel must be analyzed at the subpixel level.

The Subpixel Classifier process automatically determines representative backgrounds in the scene and subtracts of background contributions to find the best match to the input signature spectrum. Special spectral filters and matching techniques are used to determine the best-fit fractional contribution of the signature material. SEST is another spectral matching process that is fast and noise resistant. The BANDS process can be used with hyperspectral data to detect underlying features and characterize their spectral position, width, and intensity. When a material has an absorption or emission feature in its reflectance spectrum, this feature usually shows up in the mixed pixel spectrum as well and can be detected using BANDS through the presence or absence of the feature.

Once the pixel spectrum has been characterized, a number of properties about the earth's surface can be determined. Spectral processing can be used for:

· Material identification and quantification
· Anomaly detection
· Change detection
· Landcover use and mapping
· Water quality monitoring and mapping
· Bathymetry
· Submerged aquatic vegetation mapping
· Ecosystem analysis
· Environmental impact studies
· Intelligence gathering
· Intelligence preparation of the battlefield (IPB)
· Camouflage detection
· Target detection