Optimization of recognition of micro-objects based on the use of morphometric characteristics of images

Abstract

Constructive approaches, principles and methods of identification, recognition and classification of micro-objects based on the use of neural networks and mechanisms for extracting morphometric characteristics of images have been developed. Information processing technology based on obtaining images of micro-objects from a photo, video camera, digital microscope is proposed. A technique has been developed for interactive measurement of the size of micro-objects, counting, determining the structure, conducting statistical analysis, isolating and segmenting fragments, selecting informative points, recognizing and classifying images. A computational scheme for preliminary processing of images is built, including mechanisms for texture, contour segmentation, detection, and regulation of variables. Algorithms for learning neural networks with setting variables within the limits of permissible values, taking into account the properties of nonstationarity of image points, have been built. The effectiveness of learning algorithms combined with neural network dynamic models with mechanisms for regulating linear, nonlinear, compositional connections of neurons between network layers has been investigated. The study was carried out according to the criterion of the percentage of correct recognition. A software package for visualization, recognition, classification of images of pollen grains has been developed and implemented, which has been tested under conditions of a priori insufficiency, uncertainty and nonstationarity.