Page 287 - Digital Analysis of Remotely Sensed Imagery

CHAPTER 7

Spectral Image

Analysis

ince the advent of spaceborne remotely sensed data with the
launch of the first Earth Resources Technology Satellite in 1972,
Sa huge quantity of multispectral and hyperspectral satellite
data from a wide variety of sensors has accumulated so far. Although
they have considerably expanded our ability to study the Earth’s
surface, timely processing of this plethora of data is a formidable
challenge that can no longer be met by using the traditional manual
interpretation method, which is tedious, subjective, and slow. The
solution to overcoming this limitation lies in the automatic process-
ing and analysis of remote sensing data in the digital environment.
Thanks to the development of computing technology and the avail-
ability of powerful digital image analysis systems introduced in
Chap. 4, these data are now routinely analyzed digitally to derive the
desired information automatically. Computer-assisted image classifi-
cation has considerably expedited the process of studying the Earth’s
surface from satellite data. Spectral image classification, also called
information extraction, is a process of converting satellite data into
meaningful land cover information based on pixel values in an image.
Such spectral classification is accomplished either parametrically or
nonparametrically in the multispectral domain. Depending upon the
classifier, the process can be very simple or very complex, involving
a number of steps. This chapter on spectral image classification starts
with several rudimental concepts related to image classification. The
next topic is spectral distance, the fundamental decision criterion
behind spectral classification. This chapter then concentrates on
image classification itself. The two broad categories of per-pixel
image classification methods, unsupervised and supervised, are then
discussed under separate headings, and their main features are com-
pared with each other in Sec. 7.6. Following this comparison are two
sections devoted to image classification at the subpixel level and
fuzzy image classification. Finally, this chapter ends with postclassi-
fication processing undertaken to embellish the classification results.
249