Home – Home – Jornals – Advanced Search

The main ideas used in solving the problems of detecting and classifying objects by their images using neural network technologies are reviewed. The key publications devoted to the most popular ways to improve the accuracy of classification are considered. It is shown that neural network methods of object detection have achieved significant success in the last decade due to the use of convolutional technologies and practical implementation of the idea of deep learning using large visual databases. The main disadvantages, limitations, and possible directions of development of existing approaches are considered and analyzed.

Diagnosis of Parkinson's disease is an expensive procedure that includes transcranial sonography and brain tomography. In this regard, simple and accurate screening diagnostic methods are relevant. The article deals with the analysis of handwritten static drawings of spirals and meanders using machine learning methods for diagnosing Parkinson's disease based on the publicly available HandPD dataset. Fuzzy classifiers are constructed using original methods that are able to determine the presence or absence of the disease by drawing. As the Hand PD dataset is unbalanced, oversampling algorithms are used in the work. A statistical comparison of the accuracy of the applied models and methods is carried out. The ranking of features is performed.

Results of the study usingtwo neural networks to solve the problem of detection and classification of microparticles in images obtained by restoring model and experimental holograms of aerosol media with a particle density of 180000 cm^-3 are reported. The first neural network is trained on a database of manually labeled particle images, while the second neural network is trained on a database based on algorithms for automatically distributing certain parts of images into classes. The results of operation of two neural networks are compared, ways of their further improvement are presented.

Modern small and medium-sized enterprises, such as IT companies, law, consulting, and logistics firms, use many electronic tools for management purposes to ensure internal information exchange. However, there are a number of tasks, such as registration of vacations, project reporting, issuance of professional literature, etc., which do not fit into the existing decision models. Some IT companies in such cases create their own services to automate the same type of procedures, but many companies do not have software solutions for automating it. The paper presents the model and architecture of software on an open software platform of the employee appraisal service - one of the above-mentioned non-standard tasks. Using the service will allow small and medium-sized companies to apply this flexible and simple solution to frequently occurring management tasks.

We apply a polynomial approach to low-order optimal controller design for linear stationary SISO systems described by differential-algebraic equations. Critical root diagrams and root polynomials are explored in low-order controller design in classical control systems. The method is illustrated on an unstable controlled plant defined by an improper transfer fraction with a 6th-degree numerator and a 4th-degree denominator. We find the settings of a few stabilizing PI₃-controllers and select the optimal one providing the maximum relative stability; the impulse response calculation confirms that the close loop system is astatic and impulse-free. The method scheme remains the same as that for the classical control systems; however, the arising polynomial systems turn out to be of a higher degree and more difficult for numerical solving.

The possibilities of photographing objects in an aquatic environment on acoustic waves by means of circular aperture synthesis are experimentally tested. An experimental setup has been created to implement a method for reconstructing the image of objects in cross sections using a trajectory Doppler signal. Image reconstruction is carried out by scanning an object placed inside a rotating cylindrical water tank controlled by a circular synthesized aperture focus. The algorithm of focus control and image construction is given. The experiment is conducted with the simplest text objects:metal point object and metal rod with a diameter of 2 mm and a length of 10 cm. The problems of implementing circular aperture synthesis using a cylindrical rotating water tank with piezo sensors attached to the inner wall of the tank are discussed.

Given an ensemble of datasets, we study the object classification accuracy in terms of the error probability depending on the amount of processed information using various fusion schemes. Schemes of combining weak discriminant functions in each dataset as well as in an ensemble of different modality datasets are suggested. For the proposed fusion schemes, the redundancy of the error probability relative to the information-theoretic lower bound defined by the modified rate-distortion function with the Hamming distortion measure is evaluated. The experimental evaluations in datasets of signature and face images show a decrease in the error probability and its redundancy with the amount of the processed information being increased by combining weak discriminant functions.

The study deals with a conflict situation associated with the movement of clusters, including hundreds of small control objects in the field of responsibility of the radar station of an external observer. The task of the external observer is to monitor and trace the cluster movement and control it if a threatening situation occurs. The task of the cluster is to create the maximum possible uncertainty in decision making by the external observer. The cluster motion model is developed as K. Reynolds’ swarm behavior, supplemented by special coefficients. Three methods are developed for the formation of various cluster shapes, close to geometric and changing in the process of movement. A hypothesis is formulated and confirmed by mathematical modeling that, due to variations in the geometry, shape, and number of objects in a swarm cluster, there exist a possibility of creating situations where a swarm will be perceived by the external observer as a single large object different from the original one. It is shown that it is possible to create clusters simulating dangerous objects if the criterion for making a decision is the value of the effective scattering surface.

This article describes features of automated implementation encryption of both biometric data and user information resources as a whole. The created mathematical model is based on an artificial neural network designed for encryption of biometric images in the account administrator folder using mathematical methods. The object of the study is the information systems of the PC. The subject of research is the tools of protection of user information using biometric multifactor authentication.

The article is devoted to the application of machine learning methods for sentiment analysis of texts and to the study of the effectiveness of various architectures of neural networks. This direction is relevant in connection with the development of social networks and online recommendation services, where many users express their opinions about goods and services. The article presents the results of forecasting and comparing the structures of neural networks on real data from social networks. This allows determination of the most efficient architecture for the sentiment analysis of texts. The study may be useful for developers of social networks and recommendation services, as well as for researchers involved in natural language processing. The results can help improve the quality of the user opinion analysis and improve user satisfaction with services and products. Thus, the article contributes to the development of the field of machine learning and text data analysis.