Optimization of Medical Object Processing Based on Image Element Implication Matrices
Annotatsiya
Scientific and methodological principles have been developed for the optimal identification of non-stationary objects of complex structure with mechanisms for using statistical, dynamic, specific characteristics of information, as well as predictive and approximating abilities of neural networks for recognizing and classifying images of micro-objects, in particular, pollen grains, unicellular organisms in the blood. Principles for optimizing the identification and processing of images based on the use of neural networks, as well as mechanisms for combining neural networks with dynamic models, regulating the values of variables, recognizing and classifying micro-objects have been developed. A method has been developed for using statistical, dynamic, specific characteristics of images of micro-objects, as well as unique properties of neural networks. Mechanisms are proposed to ensure correct solution of problems, elimination of the problem of inaccuracy, uncertainty, loss of productivity, and consistency with real examples. Optimization of error control was carried out according to the criterion of the root-mean-square deviation, using the mechanisms of gradient optimization, least squares, heuristic search with annealing and taboos. A software package has been developed and implemented, which is specified by the number of neurons in the input layer, hidden layer, output layer, the value of the bias threshold, the same learning step, learning algorithms with direct and reverse propagation of errors at each network launch.