Bu N 
ss 
. 
(D 
ble 
ac- 
oth 
a 
and display procedures. Rectification may be geometric, in which 
the picture elements (pixels)are shifted in position; or radiometric, 
in which nonlinearites in the response of detector elements are 
compensated. Cosmetic procedures remove image defects or noise 
such as banding or sync losses that often are found in Landsat 
digital data and seen in radar or optical-mechanical scanner 
imagery. Analysis procedures are designed to increase the amount 
of information that may be extracted from the data by edge and 
contrast enhancement, or normalizing and ratioing specific bands. 
The availability of several kinds of remote sensor data in digital 
form facilitates their combination into a single image. Harris and 
Graham (1976) reported on the synergistical combination of radar 
and Landsat imagery. 
Automated Classification 
  
The key element of digital classification techniques involves a man/ 
machine interaction, whereby the analyst/interpreter will "train" the 
computer to recognize various combinations of numbers that represent 
reflectances in each of several wavelength bands for the particular 
cover types or features of interest. This training process usually 
involves data obtained over a limited geographic area. After the 
computer has been trained and the statistics defining the various 
categories of interest have been defined, the computer proceeds to 
classify the reflectance values for each resolution element in the 
entire data set. The classification algorithms range from relatively 
simplistic parallelopiped decision rules to highly camplex statis- 
tical discriminant functions based on mean spectral signatures and 
covariance between spectral channels for specific classes of information. 
Two basic approaches have been developed for training the computer 
system, as described by Hoffer (1972). The first approach is referred 
to as the "supervised technique” and involves the use of a system of 
X-Y coordinates to designate to the computer the location of known 
earth surface features. This technique has been used effectively 
for classifying and mapping agriculture areas. However, experience 
has shown that a "clustering" technique is more useful for developing 
training statistics in areas of natural cover types, such as involved 
in forestry or geologic studies. In the clustering technique, an 
entire block of data is designated to the computer and each of the 
spectral vectors contained in this set of data is automatically examined 
by the computer and the entire set of data is statistically divided 
into a number of groups or clusters, each containing data points 
having similar spectral vectors. The number of spectral groups to be 
defined is designated by the analyst. 
Forest cover maps, rangelands, agricultural soils, snow cover and other 
water resource situations, and geologic features of interest have also 
been successfully mapped using these techniques, as described by 
Hoffer (1975). 
-11-