Geoscience Reference
In-Depth Information
2.9 An Adaptive Technology to Classify and Interpret
Remote-Sensing Data of the Water Surface
Qualitatively
Collecting and processing information in a geoinformation monitoring system can
only be done by effective monitoring of the object under consideration and involves
using simulation modeling, information collecting, and information processing
(Armand et al. 1987; Burkov and Krapivin 2009).
From the position of system analysis, the system of collection and processing of
the information in geoinformation monitoring represents the structure uniting the
computers of various classes, databases and the advanced problem-oriented soft-
ware. Creation of such system demands the development of formalized description
of the information
flows and unique methodology of its processing.
Development of geoinformation monitoring systems requires the decision of a
set of problems related to the formation of data measurement
fl
fl
flows to be solved.
The problem of classi
cation of aquatories using the remote sensing measurements
is one of important among them. Various algorithms of the theory of image rec-
ognition, statistical decisions and cluster analysis are used to solve this problem.
At the present time, there are many image recognition methods, mainly because
of the variety of statements about concrete tasks. The problem of recognition
consists in the division of some group of objects into the classes at the base of
certain requirements. The objects having general properties are related to one class.
An initial data for the solution of a recognition problem are results of some
observations or the direct measurements that are named initial attributes.
Method of taxonomy (clustering) is one of the important methods of recognition
and classi
cation of images.
'
is assume set of M it is required to divide by not crossed subsets (clusters),
and the elements included in the same clusters should be close to each other enough
from the point of view of the chosen criterion of nearness, and elements from
different clusters should be far enough from each other. In one of many possible
statements of this task two numbers, a and b (0 < a < b), are given. It is considered,
that two elements x and y are close to each other enough, if p(x, y) < a, and are far
enough from each other, if p(x, y)>b.
King
Let
is method is well known in taxonomy and gives good results when the
quantity of available information in assumed clusters is moderate. According to this
method the distance between groups of points in space of attributes is de
'
ned as
distance between centers of masses of these groups. Clustering in this case is based
on the assumption that sites of the increased density in the space of attributes
correspond to similar situations.
The feature of remote measurements is information acquisition, when the data of
measurements, acquired during tracing of
flying system along routes of survey, are
directed to input of the processing system. As result the two dimensional image of
investigated object is registered. Statistical model of spottiness for investigated
space is one of models for this image.
fl
Search WWH ::
Custom Search