2687-0517 Computing, Telecommunication and Control 19 1 2026 1-115
RAR RUS 8-15 0000-0002-7060-8826 Shariaty Faridoddin shariaty3@gmail.com 0000-0003-0726-6613 Peter the Great St. Petersburg Polytechnic University Pavlov Vitalii pavlov_va@spbstu.ru
St. Petersburg, Russian Federation
 0000-0003-0473-5007 Peter the Great St. Petersburg Polytechnic University Medvedeva Ekaterina
St. Petersburg, Russian Federation
 Cross-domain deep transfer learning for branching structure segmentation Segmentation of thin, branching structures in volumetric imaging is a challenging computer vision task due to low contrast, strong class imbalance, and large variability in scale and topology. This work investigates a cross-domain deep transfer learning strategy that exploits morphological similarity between vascular-like branching patterns in different imaging modalities. Models are first pre-trained on the data-rich FIVES retinal vessel dataset and then fine-tuned on a subset of the NSCLC-Radiogenomics chest CT dataset containing annotations of branching structures. We evaluate four U-Net-based architectures – U-Net, Attention U-Net, R2 U-Net and Dense U-Net – and compare them with DeepLabV3 models using ResNet50 and ResNet101 backbones. A unified training pipeline with multi-stage intensity and contrast normalization is employed, along with a 10-fold stratified cross-validation protocol. Performance is assessed using accuracy, precision, Dice (F1 score), and area under the ROC curve (AUC). Cross-domain transfer learning leads to a substantial improvement over training from scratch: Dice scores increase from near-zero values to above 0.48 for the best-performing models. Attention U-Net achieves the highest Dice score of 0.4814, while DeepLabV3 (ResNet50) attains the highest AUC of 0.9621. Dense U-Net also provides competitive results, whereas R2 U-Net benefits less from the proposed transfer scheme. The results demonstrate that leveraging cross-domain morphological priors is an effective way to enhance segmentation of branching structures in data-scarce CT scenarios. The proposed framework provides a strong, reproducible baseline for future research on transfer learning and fine-structure segmentation in volumetric images. 10.18721/JCSTCS.19101 004.932.72 branching structure segmentation cross-domain transfer learning deep learning U-Net DeepLabV3 https://infocom.spbstu.ru/article/2026.88.1/
RAR RUS 16-25 Peter the Great St. Petersburg Polytechnic University Konstantinov Andrei andrue.konst@gmail.com
St. Petersburg, Russian Federation
 0000-0002-3796-4757 National Research University Higher School of Economics Elizarova Anastasiya
St. Petersburg, Russian Federation
 F-6480-2013 7004013271 0000-0002-5637-1420 Peter the Great St.Petersburg Polytechnic University Lev V. lev.utkin@mail.ru
Polytechnicheskaya, 29, St.Petersburg, Russia, 195251
 Enhancing Boundary Stability in Decision Trees and Random Forests: A Weighted Sample Duplication Approach Decision trees and their ensemble extensions, such as random forests, are widely used as classification models due to their simplicity and interpretability. However, in many real-world tasks where class labels overlap in the feature space, standard decision trees rely on hard splits that create fragile decision boundaries. In these regions, small perturbations in the input values can lead to misclassification, reducing the reliability of the model. To address this issue, we propose a localized data duplication mechanism that modifies the standard CART algorithm by duplicating samples located near the chosen split threshold into both child nodes. To prevent these duplicated samples from overpowering the nodes, they are assigned a reduced weight based on a smoothly decaying function relative to their distance from the threshold. This approach allows both child nodes to learn from ambiguous regions, preserving information about uncertainty while maintaining the axis-aligned deterministic structure of classical decision trees. When applied within a random forest framework, the duplication process also increases ensemble diversity. Experimental evaluation on 11 real-world datasets with varying degrees of class overlap demonstrates that the proposed modification consistently improves ROC-AUC scores and boundary stability while keeping computational costs low. 10.18721/JCSTCS.19102 004.85 machine learning decision trees random forest classification data duplication https://infocom.spbstu.ru/article/2026.88.2/
RAR RUS 26-37 Peter the Great St. Petersburg Polytechnic University Ludishchev Yaroslav ludishevyaroslav@mail.ru
St. Petersburg, Russian Federation
 Application of regularization techniques to improve forecast stability in noisy data for industrial automation The article explores modern approaches to the application of regularization methods – Ridge and LASSO – in problems of forecasting technological process parameters under industrial automation conditions. Special attention is given to addressing challenges associated with the high-dimensional feature spaces and the presence of noise in input data, which are typical in industrial environments. The theoretical foundations of these methods are presented, along with their specific characteristics and mechanisms that reduce model overfitting and enhance robustness under varying input data. An experimental evaluation of the effectiveness of regularized regression models is conducted using real industrial datasets, including time series with missing and distorted values. The results demonstrate improved forecasting accuracy, model stability, and, consequently, the reliability of automated monitoring and control systems. These methods help cope with data noise, avoid retraining, and highlight key parameters, which is especially important in conditions of limited computational resources and complex production systems. 10.18721/JCSTCS.19103 004.942 regularization industrial automation ridge regularization method LASSO industrial control systems regression models https://infocom.spbstu.ru/article/2026.88.3/
RAR RUS 38-45 0009-0004-1628-1772 Peter the Great St. Petersburg Polytechnic University Pham Huu Duc phamduc2511997@gmail.com
St. Petersburg, Russian Federation
 22735712200 Peter the Great St. Petersburg Polytechnic University Sorotsky Vladimir sorotsky@mail.spbstu.ru
St. Petersburg, Russian Federation
 Peter the Great St. Petersburg Polytechnic University Treimut Nikita treimut2013@yandex.ru
St. Petersburg, Russian Federation
 Signal distortion in polar architecture transmitters using class E RF power amplifiers The trend towards increasing the energy efficiency of transmitters used in radio and telecommunications systems focuses the attention of equipment developers on finding optimal solutions when transitioning from linear power amplifiers (PAs) to amplifiers operating in switching mode. It is shown in the paper that signal distortion level along with high energy efficiency must be considered as an important parameter of PA. Reasons of signal distortions in class E switched-mode PA used in polar architecture transmitters are given. It is shown that most sufficient distortion is caused by phase shift of the load current, which results from the nonlinear change of transistors’ output capacitance when amplifying signals with a non-constant envelope. The results of calculations of the error vector magnitude when using multilevel spectrally efficient modulation types, in particular quadrature amplitude modulation (QAM) and amplitude-phase shift keying (APSK), are presented. The results confirmed that for class E PA the most sufficient type of distortion is phase distortion. The results of analytical model are confirmed by simulation. It is shown that for 16-QAM at bit error rate = 10–4, the energy loss is 0.5 dB, while for 16-APSK its value increases to 2.2 dB. The results presented in the paper can be used in the development of signal predistortion methods for switching class E PAs, ensuring a reduction in signal distortions. 10.18721/JCSTCS.19104 621.37 switched-mode power amplifier envelope amplifier class E AM/AM AM/PM EVM https://infocom.spbstu.ru/article/2026.88.4/
RAR RUS 46-56 0009-0006-1273-6703 National Research Tomsk State University Anzhin Viktor viktor.anjin@gmail.com
Tomsk, Russian Federation
 Method for selecting the strength of unsharp masking pre-filter used to enhance the detection rate of DCT-domain image watermarks This work investigates the improvement of digital watermark detection robustness by applying image sharpening prior to the detection stage for an algorithm through the use of DCT-domain for both embedding and detection. Watermark detection is treated as a binary classification problem, enabling the use of the recall metric to evaluate the quality of detection. The recall metric is calculated on a set of test images by embedding and detecting the watermark with unsharp masking applied as a pre-filtering step. A method for selecting the optimal sharpening strength coefficient is proposed, based on maximizing the recall metric under fixed embedding and detection parameters. Computational experiments across a range of distortions demonstrate that pre-filtering with the optimal sharpening strength coefficient determined by the proposed method increases the true positive detection rate for most tested distortions without increasing the number of false positive detections. 10.18721/JCSTCS.19105 004.056.5+519.237 digital watermark discrete cosine transform unsharp masking recall content protection https://infocom.spbstu.ru/article/2026.88.5/
RAR RUS 57-64 0000-0002-1477-3181 Peter the Great St. Petersburg Polytechnic University Leontiev Evgeniy johnleon010@gmail.com
St. Petersburg, Russian Federation
 Analyses of a class G power amplifier with controlled nonlinear distortion for LTE-signal The work considers an analysis of nonlinear distortions in a class G power amplifier and an analysis of an envelope tracking power amplifier with a proposed combined envelope amplifier scheme for LTE cellular base stations. Both considered envelope amplifier schemes feature an additional shaping function to compensate for the AM-AM conversion of the power amplifier, as well as DPD correction for the AM-PM conversion of the amplifier. The analysis results showed that switching between 18 supply voltage levels in the class G power amplifier allows achieving an ACPR characteristic of the output LTE signal at an acceptable level of –50 dBc. Simulation has proved that a standard predistortion technique based on LUT or a Volterra series can be applied to a class G power amplifier at N = 18. The number of supply voltage levels can be reduced to N = 10 using the proposed combined scheme or by applying additional voltage supply filtration of the envelope amplifier. 10.18721/JCSTCS.19106 621.375.026 class G power amplifier nonlinear distortions drain efficiency DPD LTE https://infocom.spbstu.ru/article/2026.88.6/
RAR RUS 65-79 0000-0001-8591-9080 Peter the Great St. Petersburg Polytechnic University Wang Shan wangshan@mail.ru
St. Petersburg, Russian Federation
 Nikiforov Igor igor.nikiforov@gmail.com A Lyapunov-based dynamic scheduling algorithm for heterogeneous computing clusters The paper proposes a Lyapunov-based dynamic scheduling algorithm for heteroge-neous computing clusters, targeting fine-grained resource control under bursty and latency-sensitive workloads. By constructing a quadratic Lyapunov function and applying a drift-plus-penalty framework, the scheduling problem is formulated as a two-criteria optimization problem balancing queue stability and scheduling delay. A dynamic control parameter V is introduced to quantitatively regulate the trade-off between backlog stability and delay minimization. Sensitivity analysis demonstrates an O(1/V) backlog and O(V) delay trade-off. Experiments conducted on the Alibaba GPU cluster trace dataset show that under burst-dominant workloads, the proposed method reduces average scheduling delay to 0.2663 seconds, while achieving a 0.5459 resource utilization and a 0.6489 fairness index. The method is particularly suitable for latency-sensitive and dynamically fluctuating environments. 10.18721/JCSTCS.19107 004.42 Lyapunov optimization drift-plus-penalty resource scheduling cloud computing two-criteria optimization https://infocom.spbstu.ru/article/2026.88.7/
RAR RUS 80-90 AAH-8784-2019 35303230700 Peter the Great St.Petersburg Polytechnic University Vyacheslav P. shkodyrev@imop.spbstu.ru
Polytechnicheskaya, 29, St.Petersburg, 195251, Russia
 0000-0002-4685-8569 Peter the Great St. Petersburg Polytechnic University Konnikov Evgenii konnikov.evgeniy@gmail.com
St. Petersburg, Russian Federation
 Peter the Great St. Petersburg Polytechnic University Polyakov Prohor
St. Petersburg, Russian Federation
 Contextual regularization of the feature space of weakly structured data for analyzing the risk topology of complex technical systems The paper addresses the problem of eliminating sparsity and “false orthogonality” in short, weakly structured technical messages that hinder systematic analysis and modeling of the risk topology of complex technical systems. A method of contextual regularization of the feature space is proposed, which treats the enrichment of vector representations as a controlled diffusion process on a graph of joint occurrence of lemmas. The context topology is specified by a weighted adjacency matrix based on positive pointwise mutual information, and the recursive diffuser performs iterative feature propagation with depth attenuation and adaptive IDF gating, which suppresses noisy connections and amplifies diagnostically significant terms. The regularization parameter tuning is formalized as a task of maximizing the target quality functional, combining metrics of structural separability and semantic completeness with a threshold penalty for separability degradation. A priori, the limited nature of the diffusion process is demonstrated, and the elimination of orthogonality of terminologically heterogeneous descriptions in the presence of a contextual “bridge” in the graph is proven. Experimental testing on the NRC operational message corpus demonstrates a significant increase in the semantic coherence of topics while maintaining the geometric separability of clusters. The resulting regularized space improves the interpretability of the thematic structure of incidents and creates a basis for the subsequent self-organization of the risk event taxonomy and the construction of verifiable decision support contours. 10.18721/JCSTCS.19108 004.93:62-192 weakly structured data systematic risk analysis thematic modeling co-occurrence graph diffusion feature enrichment interpretable AI accident topology https://infocom.spbstu.ru/article/2026.88.8/
RAR RUS 91-102 AAH-8784-2019 35303230700 Peter the Great St.Petersburg Polytechnic University Vyacheslav P. shkodyrev@imop.spbstu.ru
Polytechnicheskaya, 29, St.Petersburg, 195251, Russia
 0000-0002-1254-0464 Peter the Great St. Petersburg Polytechnic University Rodionov Dmitry drodionov@spbstu.ru
St. Petersburg, Russian Federation
 0000-0002-4685-8569 Peter the Great St. Petersburg Polytechnic University Konnikov Evgenii konnikov.evgeniy@gmail.com
St. Petersburg, Russian Federation
 Method for classifying risk incidents based on self-organization of semantic clusters A method for automatic classification of textual descriptions of emergency risk incidents based on self-organizing semantic clustering is presented, which does not require prior data labeling. Unlike traditional approaches, the method involves a two-stage scheme, which consists of self-organization of a latent taxonomy of incidents through hierarchical thematic decomposition of the text corpus, as well as continuous classification of new messages according to their degree of belonging to all automatically selected classes at once. This transition from rigid assignment to a single class to fuzzy membership allows hybrid incidents to be decomposed into several risk factors, reflecting their mixed nature. The developed algorithm forms an interpretable and stable taxonomy of incidents that preserves the structural isolation of clusters even with a high proportion of hybrid events. Testing on the NRC data corpus showed that most messages have a dominant risk factor with significant secondary components. The average semantic consistency of clusters was ~0.62 (cosine measure), and the classification confidence is distributed around the mean, reflecting the presence of both pure and mixed incidents. The results confirm that the proposed method provides a mathematically correct decomposition of complex situations into a set of risk factors and reduces the sensitivity of classification to noise and inaccuracies in the input text. The methodology is focused on proactive risk analysis in complex technical systems and can be used for automated decision support in industrial safety systems. 10.18721/JCSTCS.19109 004.91:005.334 risk incidents unstructured data semantic analysis thematic modeling clustering fuzzy classification risk taxonomy https://infocom.spbstu.ru/article/2026.88.9/
RAR RUS 103-115 0009-0006-1822-7117 Saint Petersburg Mining University Kozhubaev Yury kozhubaev_yun@spbstu.ru
St. Petersburg, Russian Federation
 Monitoring and diagnostics of electromechanical systems based on machine learning Induction motors, widely used in electromechanical equipment of mining enterprises, are susceptible to failure due to frequent starts, overloads, and wear, leading to accidents and economic losses. Induction motors are one of the main sources of kinetic energy in industry and agriculture. Motor failure leads to shutdown of the technological process and reduced efficiency, requiring regular monitoring. Traditional diagnostic methods based on the analysis of individual signals and classic machine learning with manual feature selection are insufficiently reliable under variable operating conditions and are highly susceptible to human factor. This paper proposes an approach to diagnosing induction motor faults based on a deep residual network using signal analysis, deep and transfer learning, and information fusion. Various three-phase current input strategies are implemented, and a model capable of automatically extracting informative deep features from the current signal is constructed. The experimental results confirm that the proposed deep learning-based model provides higher diagnostic accuracy compared to traditional machine learning algorithms. 10.18721/JCSTCS.19110 622.271:621.31:004.8 motor fault diagnosis deep residual network information fusion theory machine learning induction motors https://infocom.spbstu.ru/article/2026.88.10/