Optoelectronic Information-Power Technologies

Experimental study of a determinative chaos generator based on a transistor structure

O.V. Osadchuk — 2025-06-18

The work presents an experimental study of a new circuit solution for a deterministic chaos generator based on a bipolar transistor structure with negative differential resistance. This chaos generator system has three dynamic variables: the voltage on the equivalent capacitance of the transistor structure between the collectors of the first and second bipolar transistors, and the third is the current flowing through the inductance of the oscillatory circuit. The dynamic processes of deterministic chaos are determined by the reactive properties of the transistor structure with negative differential resistance. Experimental studies were conducted from low frequencies to microwave frequencies to determine the optimal operating frequencies for various tasks of using the developed device. The I-V characteristic, the Smith chart of the S11 parameter, the S11 impedance, the active and reactive components of the impedance, the equivalent capacitance and inductance, and the SWR of the chaos generator based on two bipolar transistors in the frequency range from 15 kHz to 1 GHz were obtained. And also experimental oscillograms of the developed chaos generator were obtained. In comparison with analogues, the proposed and investigated deterministic chaos generator has improved loading capacity and higher speed, has a short time of establishment of stationary oscillations.

Development of a determinative chaos generator based on a transistor structure with negative resistance

O.V. Osadchuk — 2025-06-18

The paper proposes and investigates a new circuit solution for a deterministic chaos generator based on a bipolar transistor structure with negative differential resistance. The field of data transmission has expanded in recent years due to the rapid development of communication infrastructure, such as mobile and Internet networks. Ensuring secure data transmission in electronic communication and radio engineering devices and systems is crucial for maintaining security and realizing the full potential of these infocommunication technologies. Among the promising developments in physical-level security in data transmission is the integration of chaos theory, which increases security by using the unpredictability inherent in chaotic signals. The paper considers the possibility of obtaining a chaotic mode in a semiconductor generator based on a bipolar transistor structure with negative differential resistance. This chaos generator system has three dynamic variables: the voltage on the equivalent capacitance of the transistor structure between the collectors of the first and second bipolar transistors, and the third is the current flowing through the inductance of the oscillatory circuit. The dynamic processes of deterministic chaos are determined by the reactive properties of the transistor structure with negative differential resistance. A mathematical model of the deterministic chaos generator has been developed in the form of a system of first-order differential equations based on the state variable method, which allows determining the value of the output signal frequency depending on the supply and control voltages, as well as the parameters of the main elements of the oscillator at any point in the circuit at a given time. Using the MATLAB program package, a computer circuit engineering study of the parameters and characteristics of the generated electrical oscillations in a chaotic mode was carried out. In comparison with analogues, the proposed and investigated deterministic chaos generator has improved load capacity and higher speed, has a short time for establishing stationary oscillations.

Customer support process problems and their all-in-one resolution

R.V. Slobodian — 2025-06-18

This paper examines the critical aspects of enhancing Customer Support Systems with an aim to have them sorted out by integrating advanced computational techniques and automation. Efficient use of computational Systems across various fields, such as science, business, and engineering, relies heavily on high-quality data and sophisticated processing. Clearly organized data and well-defined tasks are essential for maximizing Customer Support System effectiveness. The study highlights that current implementations often fail to cover end-to-end scenarios comprehensively. Effective use of tools for dynamic workload management and real-time data validation presents significant challenges. Integrated solutions are needed to handle the entire lifecycle of customer support requests - from data gathering to task allocation, and finally, to managing agents' skills based on customer reviews. A holistic approach using AI and machine learning can improve task management in customer support, resulting in better data quality, efficient task distribution, and enhanced agent performance.

Improved method and tools with automatic adjustment of electrical signal parameters for detection of the reverse laryngeal nerve

M.P. Dyvak — 2025-06-18

The concept of using electromyography during thyroid gland surgery is considered. The electrophysiological features of surgical wound tissues, namely the muscle membrane potential of the vocal cord, were investigated. The analysis of EMG hardware that can be used during thyroid gland operations is carried out. The choice of EMG sensor characteristics that can be implemented in the existing complex of RLN monitoring is justified. The complex of RLN monitoring is based on a single-board computer, Raspberry Pi 4 Model B. A description of additional hardware elements to combine complex sensor and software for its functioning is provided. The developed EMG sensor was tested on a different type of low-voltage signals. It was able to detect signals and it forms 197 uV (1 Hz), 556 uV (20 Hz), and 1650 uV (10 Hz). The tests conducted show that the developed EMG sensor can detect the muscle membrane potential of the vocal cord.

RAG efficiency improvement for building intellectual scientific knownledge databases

S.V. Khruschak — 2025-06-18

The article describes the development of an intellectual knowledge base based on scientific articles using large language models in the mode of generation by augmented search. Various methods of increasing the relevance of the sample of cited sources and generated answers of the language model and the choice of approaches to building language generative systems taking into account the specifics of scientific materials in Ukrainian and English are investigated. The use of different language models for generating answers is also considered. In the course of the study, a set of criteria for a comprehensive evaluation of generative systems was selected and recommendations for building scientific intellectual knowledge bases were provided.

An intelligent agent has been developed that allows searching and analyzing scientific articles and providing document citations in a convenient interactive form.

Agile technology for developing an intelligent population development forecasting system

D.I. Uhryn — 2025-06-18

The article is devoted to the development of an intelligent population forecasting system that uses machine learning methods to analyze historical demographic data. The paper considers modern challenges of demographic development that require accurate population forecasting for effective strategic planning. The article presents a description of demographic forecasting methods, formalization and mathematical models, such as linear and polynomial regression, as well as other models that can be used for forecasting. A machine learning model generation module has been developed that automates the process of building forecasting models based on historical demographic data. Data preprocessing functionality has been implemented, including automatic filling of missing values, data normalization and anomaly detection. Machine learning algorithms have been selected and integrated, quality assessment and model optimization have been carried out, and the possibility of retraining models has been provided. An interface for integration with other information systems has been developed. The results obtained demonstrate the flexibility and effectiveness of the proposed approach and the possibility of its use in the field of strategic planning of socio-economic development.

Agile risk management methodologies in the life cycle of an intelligent system for forecasting solutions of market share dynamics

D.I. Uhryn — 2025-06-18

The article investigates the problem of forecasting market share dynamics using modern machine learning methods. The high volatility of financial markets and a significant level of uncertainty make the use of automated intelligent systems relevant for increasing forecasting accuracy and optimizing investment strategies. The proposed system combines Prophet and LSTM (Long Short-Term Memory) machine learning models for time series analysis, as well as the Monte Carlo method for risk assessment. An algorithm for collecting, cleaning, and preprocessing financial data has been developed, which includes obtaining historical stock prices from the Yahoo Finance platform, normalization, eliminating outliers, and forming training samples. The system architecture consists of modules for collecting and processing data, building forecasting models, and assessing risks. An experimental study of the effectiveness of the proposed methods based on real financial data was conducted. A comparative analysis of forecasting accuracy showed that using LSTM allows achieving an average accuracy of 92.4%, while Prophet demonstrates an accuracy of 88.7%. Risk assessment using the Monte Carlo method allowed us to determine the probability of extreme changes in asset values and their impact on the investment portfolio. The results obtained confirm the feasibility of using the proposed system for forecasting financial markets. Further research will focus on improving the accuracy of the models by integrating additional macroeconomic indicators and improving adaptive mechanisms for setting forecasting parameters.

Methodology of development and implementation of an intelligent sales forecasting information system for effective inventory management

D.I. Uhryn — 2025-06-18

The study is devoted to the development and implementation of a flexible sales forecasting methodology for efficient inventory management in stores and warehouses. The proposed model is based on machine learning methods and takes into account changing market conditions, allowing for adaptive forecast updates. The main stages of the research include analysing existing forecasting methods, selecting machine learning algorithms, developing a prototype model, and evaluating its accuracy and economic effect. To implement the model, the AutoML .NET framework was used, which provides automatic selection of the most efficient algorithms and hyperparameters. The results of model training experiments on data sets of different sizes demonstrated high forecasting accuracy using FastTree, FastForest, SDCA, and LightGBM algorithms. The effectiveness of various parameter optimisation strategies was also investigated, allowing the model to adapt to new market changes. The proposed methodology helps to reduce risks in the inventory management process, increase the efficiency of business processes and minimise costs associated with excess or shortage stocks.

Image classification using optical-digital image enhancement methods and deep learning in endoscopic examinations

Yu.Eu. Poudanien — 2025-06-18

Gastrointestinal tract (GIT) diseases remain among the most pressing challenges in modern medicine, with external environmental factors affecting human health negatively. The rapid development of artificial intelligence and computer vision is aimed at improving existing methods for disease detection through the analysis of biomedical images. This study summarizes recent scientific advances in endoscopy that integrate machine learning with both digital and opto-digital image enhancement technologies. The paper reviews sources evaluating the use of white light imaging (WLI) and various enhancement modes such as NBI, BLI, i-Scan, and FICE. A classification of endoscopic image enhancement methods is provided, along with recommendations for their application based on anatomical regions of the GIT. In addition, the study presents an overview of the use of enhanced endoscopic imaging and its combination with computer vision for increasing diagnostic parameters such as accuracy, specificity, and sensitivity based on data obtained during gastrointestinal examinations. On average, sensitivity increased by 17%, and specificity by 39% compared to results from novice endoscopists. The study also explores the trend of developing new architectural approaches for integrating opto-digital and digital methods into machine learning, as well as a comparison of diagnostic quality between AI systems and human endoscopists.

An analysis of the current state of such technologies is presented, along with prospects for the development of machine learning in automated computer-aided diagnosis (CAD) systems. Challenges related to classification accuracy degradation are identified, their causes analyzed, and recommendations for performance improvement are provided. Automated CAD systems are viewed as an effective support tool for young physicians in pathology detection, helping to reduce examination time and minimize the risk of missing critical areas that require focused attention.

Research on melanoma depth of invasion prediction method

Zhao Caifeng — 2025-06-18

Melanoma, a highly malignant skin tumor, relies on its Depth of Invasion (DoI) as a critical metric for assessing tumor malignancy, predicting patient prognosis, and guiding treatment strategies. Traditional DoI measurement methods are manual, time-consuming, and prone to errors due to complex tissue morphologies and the need for fine annotations. This study introduces a novel Convolutional Neural Network (CNN)-based framework that integrates image patch classification with morphological processing to achieve high-precision DoI prediction under coarse annotations.

The approach comprises four modules: pathology tissue differentiation using Otsu thresholding and morphological operations, lesion and epidermal region identification via EfficientNetB0 classification, and DoI measurement through least-squares boundary fitting. Experimental results on a melanoma dataset demonstrate a Mean Absolute Error (MAE) of 0.503 mm and a Root Mean Square Error (RMSE) of 0.169 mm, significantly outperforming traditional segmentation networks such as UNet and Attention-UNet. This method provides a robust and efficient solution for automated melanoma diagnosis, with substantial potential for clinical translation.

Analysis of approaches to improving intelligent parking management technologies

V.O. Kopytsia — 2025-06-18

Rapid growth in urban motorization has led to a critical shortage of parking spaces, contributing to increased congestion, higher emissions of harmful pollutants, and a decline in residents’ quality of life. This study examines contemporary methods for enhancing parking management technology, moving beyond traditional manual control and static data toward intelligent systems capable of adapting to real-time traffic dynamics and demand. The research focuses on solutions based on the Internet of Things, automated parking complexes, artificial-intelligence algorithms for occupancy forecasting, and dynamic pricing mechanisms. Methodology includes a systematic review of over forty scientific publications from 2018 to 2025, comparative analysis of technical and economic performance indicators for various technologies, SWOT analysis, and scenario modeling that incorporates social and environmental considerations. Findings indicate that deploying IoT solutions with sensor-based monitoring and mobile applications significantly reduces the average time spent searching for a parking space and corresponding CO₂ emissions. Automated parking systems deliver high vehicle density and lower operational costs, while artificial-intelligence algorithms improve the accuracy of demand forecasts. Dynamic pricing balances demand across different times of day, helping to alleviate congestion. The practical significance of this work lies in the development of recommendations for integrating these technologies into urban infrastructure and in crafting a roadmap tailored to the specific needs of Ukrainian cities. The proposed approaches can guide local authorities and investors in optimizing parking resources, enhancing urban mobility, and reducing environmental impact.

Use of neuroheadsets for diagnostics of diseases

O.N. Romaniuk — 2025-06-18

The article highlights modern approaches to the use of neuroheadsets in the diagnosis of psychoneurological diseases, including depression, anxiety disorders, epilepsy, schizophrenia, Parkinson's disease, and Alzheimer's disease. The main focus is on the registration and analysis of electroencephalographic signals, which provide a non-invasive assessment of the functional state of the brain. The significance of rhythmic activity of various frequency ranges — in particular, alpha, beta, theta, and delta waves — as markers of certain disorders is revealed. It is shown that depression typically exhibits a decrease in alpha activity in the left frontal cortex, and anxiety disorders typically exhibit an increase in high-frequency beta activity. Changes in the spectral composition of signals in epilepsy are analyzed, in particular, focal disturbances and paroxysmal complexes, which can be recorded using neuroheadsets in clinical or home conditions. The article also provides information on the reduction of coherence and variability of EEG signals in Alzheimer's disease and changes in electrical activity in patients with Parkinson's disease. Considerable attention is paid to the possibility of using neurofeedback technologies within the framework of cognitive and everyday rehabilitation, which are based on the patient's active control of their own electrophysiological reactions. The practical feasibility of using neuroheadsets for the initial screening of the patient's condition, monitoring the dynamics of treatment and assessing the effectiveness of psychotherapeutic and pharmacological approaches is emphasized. As a result, it is concluded that neuroheadsets open up new opportunities for rapid, safe and economically accessible diagnostics of nervous system disorders in a wide range of patients of different ages.

Method of segmentation of OCT images using a convulsive neural network

A.V. Shcherbatyuk — 2025-06-18

The article analyzes the methods of segmentation of optical coherence tomography images, creates a convolutional neural network model U-Net, processes a series of test images from an open database, and compares the results of processing with other algorithms using the structural similarity index (SSIM). Pre-processing of test images to improve the quality of segmentation is also performed. Preprocessing of test images was also carried out to improve the quality of segmentation. In this work, the U-Net convolutional neural network was created, trained and applied. Existing methods of segmentation of optical coherence tomography images for the diagnosis and monitoring of ophthalmic diseases were considered. The advantages of using the U-Net deep convolutional neural network in comparison with classical methods, such as the Sobel operator and the Pruitt operator, were analyzed. Unlike classical algorithms, which have limited ability to adapt to noise, image heterogeneity and pathologies, U-Net provided higher accuracy of image segmentation.

Analysis of decision support methods in polarization introscopy systems of biological tissues and fluids

V.V. Sholota — 2025-06-18

The article discusses the features of the application of decision support methods based on machine learning, fuzzy logic and neural networks in polarization introscopy systems of biological objects. It was determined that methods such as fuzzy logic, some machine learning methods (decision trees, XGBoost) and neural networks (multilayer perceptron) allow to achieve an increase in the accuracy of polarization diagnostics of BS to the level of 81-98%. However, the obtained accuracy results may be overestimated due to the imperfection of the evaluation models and methods of sample formation, which requires further research. A comparative analysis of their accuracy characteristics is presented, taking into account the input data, software implementation and the type of pathologies diagnosed in the introscopy system.

Intelligent echocardiographic image processing systems for assessing the functional state of the heart

S. Pashkovskiy — 2025-06-18

Ultrasound images of the heart are an important source of diagnostic information for the detection of cardiovascular diseases. Today, automated processing and analysis of such images are actively studied in the fields of telemedicine, digital medical image processing, and artificial intelligence, in particular, to accelerate and accurately diagnose cardiac pathologies. This paper considers a new approach to processing echocardiographic data, which involves converting ultrasound videos or series of images into color phase space projections. This allows you to create informative visual representations suitable for analysis using deep convolutional neural networks. This approach has two key advantages: [1] it provides the ability to use modern deep learning architectures for the recognition of cardiac pathologies, [2] it allows the use of transfer learning techniques, which significantly increases the efficiency of the model even on small data sets.

Intellectual method of supporting decision making in a multi-parameter system of azimuthally invariant Mueller-polarimetric in pathologies assessment

N.I. Zabolotna — 2025-06-18

The article presents a method for supporting decision-making in a multiparametric system of Muller-matrix diagnostics of biological layers based on statistical and wavelet analysis of a collection of azimuthal invariants of Muller-polarimetry and decision tree models to increase the accuracy of decisions. Training decision tree models based on minimization of the Gini index for informative features of the distributions of azimuthally independent invariants of the biological layer of the cervix are developed and the accuracy of pathology detection based on them is assessed. The experimental application of the improved PPR method in the differentiation of functional states of "normal" and "pathology" of the cervical muscle tissue of the uterine cervix with the measurement of ten distributions of azimuthal invariants of the Muller-polarimetric parameters of the uterine cervix has been demonstrated. An increase in the diagnostic accuracy of uterine cervix samples to the level of 97.2% has been achieved.

Application of artificial intelligence for automated interpretation of optical retinal images in diabetic retinopathy

O. Kornilenko — 2025-06-18

This article explores the application of artificial intelligence (AI) for the automated interpretation of optical retinal images in diabetic retinopathy. It presents the main imaging methods, including fundus photography and optical coherence tomography, and analyzes deep learning algorithms used to detect retinopathic changes. The study evaluates the effectiveness of current autonomous systems, such as IDx-DR and EyeArt, and outlines key limitations of their use. Special attention is given to ethical, technical, and legal aspects of AI implementation in ophthalmic practice. The article highlights AI’s potential as a tool for early screening and prevention of vision loss in diabetic patients.

Comparative analysis of the accuracy of classification of electromyographic signals by second-order difference graphs for differentiating types of pain in the lower back

T. Zhemchuzhkina — 2025-06-18

Low back pain is the main cause of disability. From the pathophysiological point of view, there are functional and dysfunctional types of pain. Functional pain has a specific organic cause in the form of direct tissue damage and activation of peripheral pain receptors or damage that affects the somatosensory system. Dysfunctional pain is formed as a result of neurodynamic disorders in the central nervous system. As a rule, when examining patients with dysfunctional pain, it is not possible to identify organic diseases that could explain the occurrence of the pain syndrome. This work is devoted to studying the possibilities of classification methods for differentiating functional and dysfunctional pain based on the second-order difference plot of the electromyographic signal. Electromyographic signals have extremely complex characteristics that resemble chaotic processes in nature. The second-order difference plot allows to analyze the degree of variability or chaos in a set of electromyographic data.

Analysis of methods and systems for recognition of ear pathologies on otoscopic images

A. Marchuk — 2025-06-18

An analysis of methods and tools for the analysis and classification of ear pathologies was conducted, identifying their application features, advantages, and disadvantages. As a result of this work, ways to improve ear pathology recognition systems were determined. For this study, the free software package Image Composite Editor (ICE) 2.0 (Microsoft) was used to generate seamless composite images. The combination of different methods and algorithms for image processing and classification significantly increases the reliability of the results obtained. Further studies to improve the accuracy of disease diagnosis can be aimed at combining different image processing algorithms and machine learning algorithms.

Automation of the full cycle of cryomicroscopic image processing: from collection to analysis

Yu.V. Samokhin — 2025-06-18

The article discusses the process of automating the full cycle of cryo-microscopic image processing using modern cloud technologies, annotation tools, artificial intelligence, and integration with platforms for scientific analytics. It describes a pipeline that includes stages such as data collection, storage using the MinIO cloud storage, image annotation with CVAT, the application of artificial intelligence models for inference, and result visualization. Special attention is given to the integration with Jupyter for scientific analysis and Docker to ensure scalability and reproducibility of the entire process. The advantages of automation are highlighted, providing convenience, scalability, reliability, and the ability to reuse results in scientific research, significantly enhancing the efficiency and accuracy of cryo-microscopic image analysis..

Improved model of ELASTIC NET regularization for financial time series

R.N. Kvyetnyy — 2025-06-18

This paper proposes a modification of Elastic Net regression for short-term forecasting of financial time series by introducing Gaussian weight decay. The new approach is designed to smooth the abrupt “jumps” between the last historical observation and the first forecast—an issue typical of standard regularization. To assess its effectiveness, we formally derive the Elastic Net model with four weighting schemes (no decay, linear, exponential, and Gaussian) and conduct empirical experiments on the S&P 500, Dow Jones Industrial Average, and Nasdaq Composite indices over the period 2020–2025. The results demonstrate that Gaussian decay minimizes the transition gap and achieves the lowest RMSE and Deviation for the S&P 500 and Nasdaq Composite, whereas exponential decay proves optimal for the Dow Jones Industrial Average.

Study of the double trigger phenomenon and comparison of minimax approximation with L2-regularization

M.I. Kryvosheia — 2025-06-18

This paper investigates the phenomenon of double descent and proposes the use of minimax approximation (L∞-norm) as an alternative to L2-regularization to improve the quality of model approximation. Double descent describes the dependence of the error on the complexity of the model: the error first decreases, then increases due to overfitting, and then decreases again. In contrast, in experiments with a model without regularization, a predominantly increasing trend of the error with short periods of decline was found, which is observed for an incomplete manifestation of the phenomenon. This is probably due to anomalous points in the data that caused an exponential increase in the error at high powers. Three approaches were noted: a classical model without regularization, a model with L2-regularization, and minimal approximation. L2 regularization added a penalty for large coefficient norms, which stabilized the error and prevented overfitting, especially at high polynomial degrees (200+). Minimax approximation minimized the error, thereby providing better maximum anomaly robustness and outperforming L2 regularization at low degrees (up to 50). The results confirmed that minimax approximation is more effective for problems with anomalies, while L2 regularization performs better on complex models with high polynomial degrees. The findings contribute to the understanding of the double descent phenomenon and show the practicality of applying different approaches due to data features and model requirements.

Peculiarities of associative data processing in intelligent systems

T.B. Martyniuk — 2025-06-18

Associative operations are computational massively parallel procedures over large data sets. This explains their widespread use in such application areas as database management systems (DBMS), searching and sorting IP addresses in computer networks, and ranking data, for example, in decision-making subsystems as part of intelligent systems, in particular, for medical diagnostics. This is due, not least, to the fact that associative operations include selection by foreign key, searching for data by analogy, sorting and ranking of elements of a data set. This paper presents the results of an analysis of the features of the application of associative data processing methods for solving problems in intelligent systems. The definition of intelligent memory is considered as one that is expanded due to the functional capabilities of associative memory, i.e. memory with content-addressing. In this case, associative data processing includes not only a search by association, that is, by a foreign key, but also a search for an extreme (maximum/minimum) element in a numerical array. Another example of the application of associative data processing are varieties of neural networks that perform the functions of auto- and heteroassociative memory. The use of neural networks in intelligent control systems of mobile robots is especially relevant today, since their structure is provided by associative processing levels. Another popular approach is the use of a classifier with extended functional capabilities as part of decision support subsystems for expert systems for various purposes. These examples indicate a specific connection between associative data processing methods and the implementation of neurotechnologies in the creation of intelligent systems for various purposes.

Adaptive data transmission method in information channels of telemedicine systems

P.O. Yakovyshen — 2025-06-18

The paper proposes an adaptive data transmission method for information channels in telemedicine systems, which involves dynamic compression adjustment, routing optimization, and load balancing. To validate the effectiveness of the method, computer simulations were conducted, and the results demonstrate improved transmission reliability under challenging network conditions.

The proposed approach can enhance the performance of information channels in telemedicine systems, particularly for video consultations, remote patient monitoring, and real-time transmission of diagnostic images.

Development and application of a computer program for assessing the quality of image processing based on the study of convections

M.O. Tsarenko — 2025-06-18

Cluster-based digital filters occupy a key place in computer image processing programs for adjusting the shift in sharpness, the visible border, and so on. Using the method of learning the power of such filters, you know, the beginners and the students have developed a computer program that makes it possible to scientifically, vikorista kernels of different sizes to isolate the differences in the Gortkov filters (sharpness shift, pitch, edge detection, embossing) to process the image, as well as evaluate the brightness of their work using additional criteria of peak signal to noise ratio (PSNR) compared to the original and edited images.

The program is implemented in object-oriented Java programming with the AWT and Swing libraries, which are designed for processing filters of any size in JPG, JPEG, PNG, BMP or GIF formats. The principles of operation of the convolution kernel, methods of processing noise, implementation of programs and instructions for setting the valves are described. Added functionality for entering a custom convolution kernel, processing images using Gaussian noise (σ = 25.0) and “salt-pepper” type noise (5% neutrality), with the further possibility of updating the image by resetting the noise. The program allows you to use convolution kernels with any weight coefficients. The program uses the PSNR criterion to evaluate the luminosity of image processing.Given the widespread use of convolutional filters in computer vision and digital signal processing, it is an important task to demonstrate and quantify their effectiveness.

To solve this problem, we developed a computer program that compares different convolutional filters (sharpening, blurring, edge detection, embossing, and an eigenfilter) for image processing. The quality of processing is evaluated using the peak signal-to-noise ratio (PSNR) between the original and processed image.

Denormalization techniques for IOT data warehouses: balancing query performance and data redundancy

M.V. Talakh — 2025-06-18

This article explores the impact of denormalization techniques on query performance in IoT data warehouses while maintaining acceptable data redundancy. It analyzes normalized and denormalized approaches in a smart home IoT environment using Azure Synapse. Empirical testing (10,000–5 million records) shows that strategic denormalization combined with columnar storage optimization improves performance by up to 94%. Evaluating four key optimization techniques (Join Reduction, Columnar Storage, Query Complexity Optimization, Temporal Scaling Optimization), we find that denormalization initially increases storage needs by 16% (120 GB vs. 103.5 GB), but columnar compression reduces the final storage size by 50.4% (17.1 GB vs. 34.5 GB). The study provides practical insights into balancing query performance and data redundancy in high-speed IoT environments.

Improved method of adaptive histogram equalization for color fundus images

S.A. Andrikevych — 2025-06-18

The paper investigates the improvement of the visualization quality of color fundus images using the contrast-limited adaptive histogram equalization (CLAHE) method. The method is applied to the R, G, B channels of images from the HRF database. The results showed an increase in the average contrast, and visual analysis confirmed better visibility of fundus vessels while preserving local details. The proposed approach is effective for image preprocessing in medical diagnostics. The proposed CLAHE method by separately processing the R, G, B channels has demonstrated its effectiveness in enhancing the contrast of fundus images, as evidenced by an increase in the average contrast by 4.4% and better visibility of retinal vessels, especially in the green channel, and also helps to make abnormal structures such as neoplasms or hemorrhages more visible. However, the method causes a shift in the color balance, which may affect the diagnostic value of the images, and also enhances chromatic aberration at its borders.

Agile risk management methodology for decision-making in startup projects based on stock price forecasting

D.I. Uhryn — 2025-06-18

The article is devoted to the study of the issues of risk management during decision-making in startup projects, in particular in conditions of high uncertainty and volatility of financial markets. To improve the efficiency of risk management, a method of forecasting stock prices based on modern machine learning models, such as Support Vector Regression, Random Forest and Gradient Boosting, is proposed. Experimental studies are conducted using historical financial data collected through the Yahoo Finance API, which were cleaned, normalized and supplemented with technical analysis indicators. The metrics of mean square error (MSE) and coefficient of determination (R²) are used to assess the accuracy of forecasts. The experiments have shown that the use of ensemble models and stack techniques provides high quality forecasting. Based on the results, a web application has been developed to integrate forecasts into the decision-making process in startup projects. The application allows investors and managers to analyze market trends, assess risks and make informed investment decisions. The use of the proposed system helps minimize risks and increase the stability of financial results of startup projects.