Identification of Micro-Objects Based on a Hybrid Model with a Growth Cone and Growing Axons of a Neural Network
Аннотация
Scientific and methodological foundations for the identification of micro-objects have been developed based on the mechanisms for extracting useful properties, statistical parameters, dynamic and specific characteristics, as well as the use with neural networks of statistical and dynamic models generalized. The mechanisms for determining informative points by a set of characteristic functions and entropy, correlation estimates are studied. An algorithm for optimizing the identification of micro-objects has been developed using the results of contour segmentation, a set of model variables, conditional characteristics of interconnection, and the distribution function of image points. A mechanism for identification and data processing is proposed, which is focused on the use of dynamic models combined with neural networks based on polynomial functions, splines, linear and non-linear filters. Statistical, dynamic, neural networks, hybrid models based on parabolic polynomial, interpolating parabolic and cubic splines, three-layer neural networks are implemented. Algorithms for recognition and clustering of micro-objects are studied based on the selection of informative points of the image contour on the principles of potentials. Implemented a mechanism for detecting changes in the direction of the lines of global characteristics, highlighting the contour, establishing the texture features of the image. A mechanism is proposed and implemented based on obtaining a description of the distribution of axons, growing neurons in the network. The identification software package in C++ was developed and implemented in the CUDA parallel computing environment.