MODEL AND METHOD OF A MULTI-LEVEL RECOGNITION SYSTEM CONSTRUCTION

Abstract

Purpose. To consider existing approaches to the construction of recognition systems. The task of pattern recognition can be solved step by step, if in the analysis of data, hierarchical structures are distinguished in the feature space. A method has been proposed for synthesizing a multilevel architecture of a neural network system for solving diagnostic problems. Results. The method is based on the nesting of microclusters, clusters and macroclusters in the feature space. The second condition for splitting the recognition process into stages is the decomposition of characteristics set into unifying and separating subsets. With the help of a unifying subset of features, clusters are formed within the macrocluster at the first stage. In the second stage, using a separating subset of features within clusters, microclusters are distinguished, which are groups of objects with certain properties. Practical value. A two-level recognition system, obtained as a result of synthesis based on the developed method, is logically transparent and allows increasing recognition accuracy in comparison with classical perceptrons on the same set of features. The developed model and method of synthesis of a neural network multilevel system is a simple and effective means of solving practical problems of diagnostics and image recognition.

Authors and Affiliations

I. Shevchenko, N. Shkarupa, М. Gonchar

Keywords

Related Articles

SOCIAL INNOVATION INTRODUCTION PROCESS: ESSENCE, PROBLEMS, WAYS OF EFFECTIVE IMPLEMENTATION

Purpose. Definition of the essence of social innovation introduction process, justification of the end result of this process and the disclosure of the source base of its continuous implementation, and the allocation of...

DEVELOPMENT OF METHOD FOR ESTIMATING THE AVERAGE TIME FOR PRODUCTION OF IRON ORE PELLET CONSIGNMENT

Purpose. To research the process of pellet production among the number of complex, multi-dimensional technological processes. Despite the fact that the production process is automated, pellet production control systems i...

THE STUDY OF FORMALIZATION AND DEFORMED STATE OF STEP BLANKS DURING THE ENLARGING BY PROFILED DIE

Purpose. To establish the regularities of the shape change and the deformed state of stepped hollow forgings during enlarging. Methodology. Modeling the skidding process has been performed using the finite element met...

DETERMINATION OF RATIONAL CONDITIONS FOR FORMATION OF A WATER-ICE JET FLOW WITH THE SETTING OF ITS CUTTING PROPERTIES

Purpose. To present the results of theoretical and experimental studies of the production of an icy jet by means of its intensive cooling by introducing a cryogenic liquid (nitrogen) into the mixing chamber. Methodology....

OPTIMIZATION MODEL OF CHOOSING TECHNICAL MEANS OF CONTAINER TRANSPORTATIONS AND THEIR RATIONAL COMBINATION

Purpose. Тo develop a methodology of using the mathematical model that presupposes an optimization procedure with the help of which effective types of technical means and their number and structure necessary for carrying...

Download PDF file
  • EP ID EP660616
  • DOI -
  • Views 93
  • Downloads 0

How To Cite

I. Shevchenko, N. Shkarupa, М. Gonchar (2017). MODEL AND METHOD OF A MULTI-LEVEL RECOGNITION SYSTEM CONSTRUCTION. Вісник Кременчуцького національного університету імені Михайла Остроградського, 1(107), 54-63. https://europub.co.uk/articles/-A-660616