Effective recognition of pollen grains based on parametric adaptation of the image identification model

Abstract

Constructive approaches, principles and methods of image processing in conditions of a priori insufficiency, parametric uncertainty, low accuracy of information have been developed on the basis of statistical, dynamic, neural network models for identifying micro-objects. Methods of point and nonlinear verification of the correspondence of the contours of the input and reference objects based on the selection, segmentation, interpolation of specific characteristics of images of pollen grains and regulation of raster parameters are proposed. Modified computational schemes for learning neural networks with forward and backward propagation of errors, regulation of interneural connections in layers, neuron weights, variable activation functions, network architecture, training samples, superposition of continuous input-output dependencies. A computer complex for identification in the C++ language in the “CUDA” parallel computing environment has been implemented for the recognition, classification and systematization of pollen grains [1].