No 1 (2024)

The paper is devoted to the discussion of the issues of the accelerated development of the new generation National Research Computer Network (NIKS) as a single industry information and telecommunications network in the sphere of science and higher education of the country. The developed set of organizational and technical measures for the development of NIKS is being implemented as part of the national project "Science and Universities" for 2021-2024. The paper collects and systematizes key information on the progress in the fulfillment of the NIKS development project, presents and discusses the main activities implemented during the first two years of work on the national project (2021-2022), as well as plans for 2023-2024. The main attention is paid to the aspects of expanding the territorial accessibility and increasing the throughput of the backbone infrastructure, methods of selecting and connecting leading research and educational organizations of higher education to NIKS, the dynamics of the growth in the number of users is indicated, the map of the territorial coverage of the network is given and commented on in the context of subjects.

Human behavior in stressful situations depends on the psychotype, socialization on a host of other factors. Phone scammers build their conversation focusing on the behavior of a certain category of people. Previously, a person is introduced into a state of acute stress, in which his further behavior to one degree or another can be manipulated. We have developed a modification of the WFT capsular neural network 2D-CapsNet, which allowed using the photoplethysmogram (PPG) graph to identify the state of panic-stupor with an accuracy of 82%, which does not allow him to make logically sound decisions. When synchronizing a smart bracelet with a smartphone, the method allows real-time tracking of such states, which makes it possible to respond to a call from a telephone scammer during a conversation with a subscriber.

The article describes the development of the software module, which allows determining the size and volume of pulmonary nodules detected during low-dose computed tomography of the chest organs. The main focus of the article is on automatic quantification of nodules in accordance with the guidelines for the management of pulmonary nodules of the British Thoracic Society, the Fleischner Society, the Lung-RADS and European position statement on lung cancer screening. The approach presented is based on classical image processing methods and methods based on the use of neural networks (U-Net architecture). The input data are masks being obtained as a result of segmentation (it can be performed by a radiologist manually, in automatic or semi-automatic mode) of low-dose CT scans. The output data are DICOM image files being obtained from the original low-dose CT slice files by overlaying metadata, pulmonary nodule contours, long and short axes, and their lengths in mm, and a structured report (DICOM SR) containing lung nodule data in an easy-to-read format. An algorithm for calculating the position of the nodules relative to the lungs is also implemented for the possibility of comparing two studies of one patient in order to estimate the volume doubling time of the pulmonary nodules. The module being described is the part of the medical decision support system, that is being developed to solve the problems of reducing the heavy workload of radiologists and improve the accuracy of diagnosis of various diseases through the analysis of medical images.

The article discusses various image recognition technologies and proposes methods to enhance them by exploring additional features. In particular, a new approach is introduced that contributes to improving image recognition by using Harris corners as additional features in images. This significantly enhances the accuracy of the recognition classification model. The significance of this approach lies in its ability to enhance the recognition system's capabilities in detecting and highlighting key object features, ultimately leading to more reliable and efficient results in data analysis, processing, and classification. It also increases the model's robustness. Thanks to these improvements, this image recognition technology can be successfully applied in various fields where high accuracy and reliability are required in information recognition, such as medicine, vehicle classification, and more.

The article discusses the possibility of classification of images of Firing Pin Impressions with the use of Convolutional Neural Network (CNN). The aim of the work is to investigate the effectiveness of CNN for multiclass classification of Firing Pin impressions for several firearms. The scientific novelty of the research is in the development of the CNN for the classification of Firing Pin Impressions under the condition of a small number of source objects used for the CNN training (only 4 images for each class). In order to prove the effectiveness of the CNN training the augmented training database was prepared. For this purpose, each source image in the training database was cloned and eight new images with limited modifications were made. The results of the examination of developed CNN with the database including 40 different classes (firearms) show that the accuracy is about 93% if only one maximal result is considered. In case of considering three highest results, the accuracy increases to 97-98%. The presented work can be of interest for developers of software for automatic ballistic identification system and for firearms examiners of regional forensic ballistic laboratories working with digital microscopes. 

A method has been developed for selecting base modules for generating modular number systems and modular arithmetic, in which the calculation of the positional characteristic of the modular representation of a numerical quantity, which is a nonlinear function of many variables, is performed with linear complexity from the number of bases of the modular number system when calculated in the range of a single base of the modular system. This significantly reduces the bit depth (hence the amount of hardware) of additional modular processor blocks. Modular algorithmics previously lacked methods for calculating positional characteristics for such parameters. All non-modular (not parallelizable in modular arithmetic) operations of a specialized processor with multiple processor elements (data streams) and a single instruction stream are based on the calculation of positional characteristics. For numerical quantities in modular data formats, the method allows them to be performed with minimal linear complexity. The article substantiates the formulation of the problem and the goal of fast calculation of positional features in modular data encoding. A new method is described and justified. The results of numerical modeling of the method and examples of modular number systems allowing its use are presented. The analysis is given the obtained of new results.