Page 288 - Digital Analysis of Remotely Sensed Imagery

250 Cha pte r Se v e n

7.1 General Knowledge of Image Classification

7.1.1 Requirements
Although image classification is mostly performed automatically by
the computer in the digital environment, human intervention, either
prior to the classification or during postclassification, still plays an
indispensable role in its success, even though this intervention is
reduced markedly in comparison with manual interpretation. The
successful completion of image classification is impossible to achieve
without the analyst fulfilling the following three requirements.
First and foremost, the analyst must be familiar with the subject
area under investigation. For instance, vegetation mapping at the
species level is best done by a botanist. If not feasible, at least the
analyst should be knowledgeable in botany in order to produce a rea-
sonable and convincing vegetation map.
Secondly, the analyst must be familiar with the geographic area
under study, including its physiographic settings and the background
information related to the theme under investigation. Such a familiar-
ity may be gained through selective reconnaissance trips augmented
by a study of large-scale aerial photographs or other satellite images
of a larger scale. Other useful ancillary materials include the most
recent topographic maps and thematic maps. These collateral materi-
als are essential to achieving reliable classification results. The more
preparatory work done to become familiar with the study area and
the subject, the easier the task of image classification, and the more
accurate the results will be.
Finally, the analyst must have a sound understanding of the
remotely sensed data being used, such as their spatial resolution,
number of spectral bands and their wavelength ranges, and time and
date of acquisition. In addition, the analyst must possess rudimental
photo interpretation skills. Command of basic skills in using photo
elements eases the task of image analysis, and is conducive to genera-
tion of accurate results. These photo elements are critical to the proper
interpretation of aerial photographs by human interpreters. Although
the use of these elements is reduced considerably in the digital analy-
sis context, their skilful application facilitates the achievement of
more accurate classification results, especially if the classification
method is supervised.

7.1.2 Image Elements
There are a total of seven photo elements (Table 7.1): tone/color,
shape, size, shadow, texture, pattern, and location/association. Tone
refers to the darkness or brightness of a pixel. In the digital environ-
ment, tone is equivalent to gray level, representing the amount of
radiation from the scene received at the sensor. Gray level is rendered