Urzhumov D.V., Krevetskii A.V. The study of sufficient statistics of distinguishing group of point objects with a chain and cloud structures by the form of their hierarchical grouping graphs

Abstract: the article describes a parameterization of models of radar images of groups of “string” and “cluster” point objects. The authors study probabilistic characteristics of sufficient statistics for their distinction, necessary for the selection of decision rules optimal according to specified criteria under different conditions of observation. The article reviews a method of modeling the distortion of observed reference chains that reduced all the diversity of observations down to two parameters: curvature of the trajectory of the chain and degree of deviations of observed coordinates of point objects from their reference position. This technique allows formalizing and reducing the complexity of comparing competing methods of identification chains. The paper lists features of software for testing competing algorithms of distinguish group of objects. The authors suggest using ratio of the diameter of the graph of hierarchical grouping of objects detected to the total length of the edges of this graph as the statistics of distinguishing. Based on these mathematical models by statistical experiments the authors obtained sample estimates of the laws of probability for distribution of the sufficient statistic for distinguishing different values of model parameters. Taking into account the complexity and the properties of the mentioned method of distinguishing chains and groups the authors recommend it to be used in the construction of systems of recognition of group of point objects in high priori uncertainty regarding the parameters of the conditions of observation for groups with power not less than 10. The proposed architecture of the software package allows testing detection algorithms with signature having various number and type of parameters.

Keywords: