A.P. Belobrov, A.A. Kuzmin, S.A. Filist
The most important step of process of classification and identification of living systems is the step of extraction of informative features. The task of step is the generating the space of informative features where the vector of informative features would uniquely describe the object or its specific property. The use of primary space of informative features often does not lead to necessary effect. That is caused by noisiness of the input information and unacceptable initial data levels data which lead to generation of noise in decision model.
Considering synchronism of rhythms in compound system it is possible to extract the leading rhythm from the whole set of rhythms and to measure all other rhythms in units equal to the period of the leading rhythm.
The using of this approach for the analysis of electrocardiograms of patients with coronary heart disease is consi-dered as an example.
Sample electrocardiosignal is segmented on cardiocycles at the first step. The received set of segments is trans-formed to a matrix. Then the short-time Fourier transform (STFT) of rows of received matrix is carried out. If there is variability of spectral factors from cardiocycle to another cardiocycle then the spectral factor will diffuse at all rows otherwise it concentrates in the first (zero) row. Therefore variability of each spectral factor can be defined on basis of the ratio of spectral capacity of this factor in the first (zero) row to spectral capacity of the whole column.
STFT is defined in every row of the matrix of spectral factors that is received at the third step. It is done to define variability of only that factors which variability is high.
Informative signs are synthesized with the help of morphological analysis of received multidimensional frequency presentations.
The testing of the method has been done by means of neuronet structures realized in MATLAB.