The theoretical bases of the synthesis and analysis of non-parametrical models of the recognition of images of a collective type are developed. Their construction assumes the formation of a family of simplified approximations of the equation of a dividing surface relating the system of «reference» situations from the training sample with their subsequent unification in the uniform decision function with the help of the methods of non-parametrical statistics. The technique of the quantitative estimation of the area of competence of the investigated class of systems at the uniform law of the distribution of «reference» situations on their arbitaries of the density of probability is generalized.
Using the statistical estimations of the integrated parameters of the conditions of classification and the ratio «quantity of elements of the collective / volume of the training sample» the researchers have offered the criteria of the choice of non-parametrical algorithms of the recognition of images, that provide their systematization and the automation of the designing. On the basis of the randomized and iterative procedures of the formation of simplified approximations of the decision function an original combined algorithm of the optimal synthesis of the structure of non-parametrical systems of the recognition of images of a collective type is developed. This algorithm allows to double or treble their computational efficiency.
The software of the construction of non-parametrical systems of classification of a collective type is created. Their properties at the finite volumes of the training samples are researched by means of a statistical modeling method. The received scientific results are realized at a smart information technology of the automation and designing of non-parametrical collectives in the environment Delphi for the computers of a Pentium type. The results are also presented in the monography «Non-parametrical collectives of decisive rules».
The RSSR grant №98–06–12001 «Development of smart information system of complex research of the development of health of a man and the population of a region on the data of population checkouts».
The information system of forecasting and optimization of the dynamics of the parameters of health of the population of a region on the coordinated data of short time series of control of the parameters of the investigated process is created. The similar conditions are characteristic for some unique medical and biological systems. Such systems include, for example, the development of health of homogeneous groups of the population in view of ecological factors, and the processes of treatment and adaptation of a man. A new class of non-parametrical models of time medical and biological dependences based on the method of collective estimation is offered.
The non-parametrical collectives of the decision rules allow to use not only information contained in the elements of a time series of the meanings of medical and biological parameters, they also allow to open their integrated properties by means of a controlled combination of the advantages of parametrical and local approximations of the restored dependence.
With the help of non-parametrical collectives the technique of the estimation of risk factors contribution into the formation of the meanings of parameters of a sicknes rate of the population of a region is developed. The offered approach uses an opportunity of the representation of non-parametrical collectives at the linear simplified approximations in the form of a linear polynomial in the space of arguments with the nonlinear factors. This aproach allows their meanings to carry out the differentiation of the risk factors of diseases in particularly current conditions.
The information subsystem of optimization of the process of the development of health of the population of a region is developed on the basis of a statistical model and numerical methods of the search of the extremum. The purpose of the problem is to define such meanings of risk factors (in particular, ecological parameters) at which a sickness rate parameter would reach the level, given by a user — expert. This problem is opposite to the forecast of sickness rate of the population and is solved on the basis of the models of its dynamics.
The information system is introduced at the Centre of State Epidemic Inspection of Krasnoyarsk Region for the forecasting of the sickness rate of the population of the region, for the estimation of the contribution of separate ecological parameters into the change of its dynamics, the choice of favorable ecological conditions.
The information means of the estimation of the condition of functional subsystems of a man is introduced at Institute of Medical Problems of the North SB RAMS and Krasnoyarsk Hospital of Ambulance when solving the problems of the forecast of outcomes and danger of a postoperative period of the patients with a heart and pericardium wound.
The demonstration variant of the information system is developed for the users. The software of the system is aimed for working in the operational environment Windows95 on the computer 486DX4 or Pentium with the volume of operative memory not less than 8 Mb.