Remote Sensing: Classification Techniques

1
Remote Sensing Classification Techniques
2
(No Transcript)
3
Classification This is a process which groups
homogenous pixels together based on a decision
rule. Classification is normally done by a
computer system (ERDAS Imagine or Visual Learning
Systems) using mathematical techniques.
4
(No Transcript)
5
(No Transcript)
6
Supervised Classification This process uses
training data to classify unknown
pixels. Training Data are areas selected by the
user from a photo. These areas are defined by
the user and are used by the computer to
determine the classification of other pixels.
7
(No Transcript)
8
(No Transcript)
9
Unsupervised Classification This process takes
random samples within the photo and breaks the
classification into clusters. Cluster classes
are the groups that will be uses to sample the
population statistics.
10
(No Transcript)
11

• Classifiers
• Parallelpiped Classifier
• Decision Tree Classifier
• Minimum Distance Classifier
• Maximum Likelihood Classifier
• Fuzzy Classifier

12
Parallelpiped Classifier Decision is based on
highest and lowest value in each class.
13
Decision Tree Method hierarchically based
classifier which compares the data with a range
of properly selected features Features often used
are as follows.(1) Spectral values(2) An index
which is computed from spectral values. For
example, the vegetation index is a popular
indices.(3) any arithmetic value such as
addition, subtraction or ratioing.(4) Principal
components.
14
(No Transcript)
15

• Minimum Distance Classifier used to classify
  unknown image data to classes which minimize the
  distance between the image data and the class in
  multi-feature space.
• Three Distance
• Euclidian Distance
• Normalized Euclidian Distance
• Mahalanobis distance

16
(No Transcript)
17
Maximum Likelihood Classifier one of the most
popular methods of classification in remote
sensing, in which a pixel with the maximum
likelihood is classified into the corresponding
class.
18
(No Transcript)
19
Fuzzy Theory In remote sensing it is often not
easy to delineate the boundary between two
different classes. For example, there are
transitive vegetation or mixed vegetation between
forest and grass land. In such cases as unclearly
defined class boundaries, Fuzzy set theory can be
usefully applied, in a qualitative sense.
20
Expert System problem solving system which
supports expert knowledge in a computer based
system. 1) Knowledge about image
analysisProcedures for image analysis can be
made only with adequate knowledge about image
processing and analysis. A feedback system should
be introduced for checking and evaluating the
objectives and the results. (2) Knowledge about
the objects to be analyzed Knowledge about the
objects to be recognized or classified should be
introduced in addition to the ordinary
classification method. The fact that forest does
not exist over 3,000 meters above sea level, is
one example of the type of knowledge that can be
introduced
21
Land cover mapping is one of the most important
and typical applications of remote sensing data.
Land cover corresponds to the physical condition
of the ground surface, for example, forest,
grassland, concrete pavement etc., while land use
reflects human activities such as the use of the
land, for example, industrial zones, residential
zones, agricultural fields etc
22

• The classification was carried out as follows.
• Geometric correction
• A geo-coded Landsat image was produced.
• b. Collection of the ground truth data A ground
  investigation was made to identify each land
  cover class on the geo-code Landsat image as well
  as on topographic maps.
• c. Classification by Maximum Likelihood Method
• The Maximum Likelihood Method was adopted using
  the training samples obtained from the ground
  truth.

23
The land cover change can also be divided into
two a. seasonal changeagricultural lands and
deciduous forests change seasonally b. annual
changeland cover or land use changes, which are
real changes, for example deforested areas or
newly built towns.