|
Algorithms and Software
Classification Algorithm of Group Point Objects with Unordered Elements based on Closeness Probability Measure
A. Yu. Kaplina, A. A. Korotina, A. V. Nazarovb, V. L. Yakimovb a Joint Stock Venture «Radioavionika»
b Mozhaisky Military Space Academy
Abstract:
The paper presents a classification algorithm of group point objects (GPO) based on the comparative analysis of fragments of distorted images and the GPO templates. The sequences of the GPO elements of different lengths are used as fragments. The paired and angular interdot distances are used as classification signs. The probability measure of closeness, set by the expert by means of the membership function and the distribution law of probability of discrete values of classified objects signs, is used in solving a classification task.
The algorithm includes the following stages: search and comparison of fragments composition of distorted images and the GPO templates; formation of a probable assessment of closeness of GPO distorted image and each template in space of the considered signs according to the analysis of each fragment; accumulation of the received probabilities on the basis of analysis results of all distorted image fragments; ranging of the received probabilities of classifying the distorted image as the GPO template; determination of the most probable template. The algorithm provides the possibility of specifying a GPO distorted image class using logical rules and analytical expressions of the considered data domain. The example and results of the algorithm application for solving a classification task of real GPO on the basis of the analysis of their fragments in the form of sequences from two and three elements are given.
Keywords:
group point object; classification; probability measure of closeness.
Citation:
A. Yu. Kaplin, A. A. Korotin, A. V. Nazarov, V. L. Yakimov, “Classification Algorithm of Group Point Objects with Unordered Elements based on Closeness Probability Measure”, Tr. SPIIRAN, 48 (2016), 214–232
Linking options:
https://www.mathnet.ru/eng/trspy911 https://www.mathnet.ru/eng/trspy/v48/p214
|
Statistics & downloads: |
Abstract page: | 124 | Full-text PDF : | 103 |
|