Solving the problem of object classification in an image using computer vision methods in various areas of human activity

Abstract

This review article presents an analysis of the main areas of research on the topic of classification of objects in an image using computer vision methods. Computer vision methods allow you to automate the process of extracting semantic meaning from images. The classification of objects in an image means the localization of objects of interest to the researcher and their correlation with a certain class. The relevance of this topic is enshrined in the state program: the national strategy for the development of artificial intelligence for the period until 2030. The article also provides statistics on the publication activity of scientific authors on the topic “computer vision”, which also shows the relevance of this area. The work has the following structure: in the introduction of the article various statistics are given that reflect the relevance of the topic. The following is an overview of scientific research devoted to solving the applied aspects of the problem of classifying objects in an image in various fields of human activity. The main focus is on the following application areas: medicine, industry, security, transport and military affairs. The following is an analysis of the methods that are used to solve the problem of classifying objects in an image. The author distinguishes two groups of methods: classical and neural network methods. Classical algorithms and methods mean an approach to solving the problem of classifying objects in an image that does not use artificial neural networks. Conclusions. The research topic is relevant today, which is confirmed by statistics and government programs. The following disadvantages have been identified for classical methods: each new applied problem requires the construction of an algorithm for solving it, the complexity of identifying significant features and instability when working with certain types of data. For neural network methods, the main disadvantage is the dependence of the final model on the quality of the data set on which it is trained.