#4 / 2021

We present the results of studies of introduced species of the genus Grindelia: G. integrifolia DC., G. robusta and G. squarrosa (Pursh) Duna in the foothills of Adyghea. For the first time, the physiological characteristics of these species have been studied and the indicators of the water regime, photosynthesis and respiration have been determined using generally accepted research methods. Their comparative characteristics have been given, which makes it possible to assess their prospects. The studied species are defined as drought-resistant and decorative, and are of interest in the creation of collection sites for medicinal and essential oil plants.

doi: 10.53598/2410-3225-2021-4-291-46-51

46-51.pdf

(320 Kb)

Mathematics

In the work nonlinear autonomous second-order differential equations with right-hand side random processes are considered. Sufficient conditions for stochastic asymptotic stability in the large of solutions of the equations are obtained. The probabilistic estimations of the remaining of a random trajectory in a bounded region of the phase plane are given. As an example harmonic oscillator perturbed by random process is considered.

doi: 10.53598/2410-3225-2021-4-291-11-23

11-23.pdf

(501 Kb)

We show, that a continuous piecewise-smooth vector field in the plane, which has only hyperbolic singular points, that do not lie on the switching lines, only hyperbolic closed trajectories transverse to the switching lines, and does not have trajectories going from saddle to saddle, is structurally stable. The set of such fields is open and everywhere dense in the space of all continuous piecewise smooth vector fields.

doi: 10.53598/2410-3225-2021-4-291-24-29

24-29.pdf

(333 Kb)

The work considers application of index theory in qualitative theory of polynomial differential systems on plane. For example, we show that a cubic differential system cannot have three four-separatrix saddles at the equator of the Poincaré sphere. In addition, the sum of the indices of all infinitely distant equilibrium states of this system belongs to the number segment [–2, 4]. We give examples of specific systems.

doi: 10.53598/2410-3225-2021-4-291-30-45

30-45.pdf

(487 Kb)

Technical Sciences

The paper presents the results of a study of the harmful effects on images and proposes a protection algorithm for computer vision systems. An intelligent protection and pattern recognition system, based on two types of neural networks: an autoencoder and a convolutional neural network, is proposed.

doi: 10.53598 / 2410-3225-2021-4-291-52-57

52-57.pdf

(735 Kb)

The relevance of the tasks of digitalization of the activities of enterprises is increasing along with the expansion of the capabilities of computer processing of information and intelligent technologies. In the field of the timber and pipe industry, an important task of digitalization is to automate the measurement of the geometric parameters of objects, which makes it possible to speed up the processes of information exchange during the implementation of warehouse or trade operations. For a long time, the automation of the measurement of geometric parameters was carried out by deploying complex software and hardware systems, the operation of which required the involvement of personnel of appropriate qualifications. As a result, the benefits from the acceleration of information exchange and a decrease in the proportion of errors caused by the human factor were offset by the increased cost of performing business processes. Thanks to the development of computer vision technologies, photogrammetry methods, cloud technologies and mobile computing, it became possible to reduce the requirements for the complexity of the software and hardware of measuring systems and the personnel operating these systems. This paper presents the results of fundamental research devoted to the development of theoretical foundations for the construction of photogrammetric measurement systems that are invariant to photo-recording devices. The results obtained in the course of the study include a methodology for preparing input data for measurement, a methodology for the dynamic distribution of tasks across mobile and cloud subsystems, a method for calculating values based on CBR.

doi: 10.53598 / 2410-3225-2021-4-291-58-69

58-69.pdf

(387 Kb)

The paper considers the formalization of algorithms for multitasking artificial intelligence using a vector criterion. The work shows problems that arise when compiling training algorithms for multitasking artificial intelligence. Methods of solving problems are presented, including constructing self-learning neural modules, using special types of neural networks based on a vector criterion and formalizing algorithms for processing large data using a vector criterion.

doi: 10.53598 / 2410-3225-2021-4-291-70-74

70-74.pdf

(421 Kb)

The paper analyzes the problems of protecting voice biometric data. A mechanism for registration and verification of a person in a biometric system using voice data has been devel-oped. An algorithm for the functioning of the human and synthesized voice recognition system is proposed. An intelligent system has been developed that confirms the relevance of ongoing research in this area.

doi: 10.53598 / 2410-3225-2021-4-291-75-81

75-81.pdf

(913 Kb)

Deep learning is successfully used to solve various complex tasks, ranging from big data analysis to computer vision and providing information for decision makers. However, advances in deep learning can also be used to create software that poses threats to privacy and security. An example of such a recently appeared application based on deep learning is deepfake. The algorithms implemented in the application can create fake images and videos that are almost impossible to distinguish from genuine ones. In this regard, the development of technologies that can automatically detect and evaluate the integrity of digital visual media is relevant. The article provides an overview of the algorithms used to create fake images synthesized using artificial intelligence, and, more importantly, the methods proposed in the literature for detecting these images to date. We present extensive discussions on issues, research trends and trends related to deepfake technologies. Based on the analysis of the background and modern approaches to detecting fakes, the article provides an overview of the methods used to create fake images, which contributes to the development of new and more reliable means to combat them.

doi: 10.53598 / 2410-3225-2021-4-291-82-94

82-94.pdf

(527 Kb)

The purpose of this work is to conduct a comparative analysis of various methods of natural language processing to identify the most effective mechanisms for detecting the vector of social engineering attacks. The study shows that the use of the simplest language models, such as bag of words and TF-IDF, does not fully meet the requirements of the task. The use of the most commonly used word embedding models at the moment (Word2Vec, GloVe, fast Text) can give good results, but only when using a large training sample. It was also revealed that convolutional and recurrent neural networks can show good results, which can be used as a basis for the implementation of the Attention mechanism. However, the most productive will be the use of modern natural language processing methods based on the transformer architecture, especially the BERT and XLNet models.

doi: 10.53598/2410-3225-2021-4-291-95-108

95-108.pdf

(729 Kb)

Digitalization in the field of state and municipal procurement and the active introduction of electronic tools are due to the constant increase in the potential of electronic procurement (e-procurement). Special attention is paid to the unified information system (UIS) – a digital platform that provides information interaction between the subjects of this sphere. The paper examines the international experience in the digitization of the public procurement system, analyzes the functionality of the UIS for compliance with the features of the state digital platform.

doi: 10.53598 / 2410-3225-2021-4-291-109-117

109-117.pdf

(324 Kb)

Based on the analysis of existing modern social networks, the paper discusses the stages of developing the client side of the user interface of the prototype of the new online platform “NaSozvone”, which implements such social network functions as messaging, communication, dating, creating social relationships between people who have common interests related to work, music, video.

doi: 10.53598 / 2410-3225-2021-4-291-118-124

118-124.pdf

(483 Kb)

The paper deals with the development of algorithmic provision of digital support of ecologically balanced precision agriculture in the Republic of Adyghea. The publication describes basic techniques for calculating geomorphological parameters and insolation of a virtual field model. The authors develop the algorithms for calculating slope, exposure and insolation for elementary surfaces - structural elements of the model.

doi: 10.53598 / 2410-3225-2021-4-291-125-134

125-134.pdf

(379 Kb)

The paper discusses the automation of measuring the electrical characteristics of a thermoelectric converter, Peltier element, based on the Arduino hardware and software platform. The authors describe the emerging difficulties associated with the physical features of the Peltier element and the Arduino potentialities.

doi: 10.53598 / 2410-3225-2021-4-291-135-146

135-146.pdf

(486 Kb)

Memory