Review of the application of machine learning algorithms to the problem of classification of pollen grains

Abstract

The article provides an analytical review of recent world experience in the application of machine learning algorithms to the problem of pollen grain recognition. It briefly describes what characteristics are used to train machine learning models for a given task. The concept of «Computationally motivated biology» is introduced, a field in which biology is studied to model biological systems using computer science. To do this, researchers analyze the behavior of a biological system and then create tasks as an artificial model to make it easier for humans. Currently, automatic classification for pollen identification is becoming a very active area of research. The article substantiates the task of automating the classification of pollen grains. The paper mainly analyzes the latest research on the use of neural networks of various configurations for the classification of pollen grains. The results of applying various neural networks to the problem of classifying pollen grains (LeNet, AlexNet, DenseNet, DenseNet-201, ResNet-50) are considered. CNN-based methods for identifying pollen on microscope slides were analyzed and showed promising results even in the presence of fungal spores, bubbles, debris, and dust. Convolutional neural networks are presented that processed scattering and fluorescence signals from pollen grains. The fluorescence spectrum was processed using a multilayer perceptron. The method of automated clustering of pollen grains is considered, which gave promising results. A comparative analysis of the currently existing databases of pollen grains (Duller’s Pollen Dataset, Pollen 23E, Pollen73S, Pollen 13K) was carried out. The results of the competition to automate the process of classifying pollen grains Pollen Grain Classification Challenge are presented and analyzed.