Кафедра авіоніки та систем управління

Permanent URI for this communityhttp://er.nau.edu.ua/handle/NAU/58724

News

Відповідальний за розділ: Провідний фахівець кафедри авіоніки та систем управління Шугалєй Людмила Петрівна. E-mail: shugaley.lyudmyla@npp.kai.edu.ua

Browse

Now showing 1 - 12 of 12

Comparative Analysis of Text Vectorization Methods
(National Aviation University, 2023-06-27) Sineglazov, V. M.; Синєглазов, Віктор Михайлович; Savenko I. M., I. M.; Савенко, Ілля Михайлович
The paper considers methods of vectorization of textual properties of natural language in the context of the task of intellectual text analysis. The most common methods of statistical analysis of feature extraction and methods that taking into account the context are analyzed. The work describes the above types of text embeddings and their most common variations and implementations. Their comparative analysis was performed, which showed the relationship between the type of task of intellectual text analysis and the method showing the best metrics. The topology of the neural network, which is the basis for solving the problem and obtaining metrics, is described, and implemented. The comparative analysis was carried out using the relative time analysis of the theory of algorithms and classification metrics: accuracy, f1-score, precision, recall. The classification metrics are taken from the results of building a neural network model using the described framing methods. As a result, in the task of analyzing the tonality of the text, the statistical method of framing based on n-grams of character sequences turned out to be the best.
Determination of Characteristics of Infectious Endocarditis Based on Intelligent Processing of Ultrasonic Images
(National Aviation University, 2022-12-27) Sineglazov, V. M.; Синєглазов, Віктор Михайлович; Chumachenko, O. I.; Чумаченко, Олена Іллівна; Kolomoiets, S. O.; Коломоєць, Сергій Олексійович
The paper presents the pathogenetic factors in the development of infective endocarditis and identifies its predictors. The need for an echographic study associated with the search for the anatomical characteristics of infective endocarditis is shown: vegetation, destructive lesions (valve aneurysms, perforation or prolapse, etc.), the presence of abscesses, in the case of a prosthesis, a new divergence of the valve prosthesis may be a characteristic feature. A classification of research methods is presented that includes classical approaches of echocardiography (transthoracic, transesophageal) and new multidetector computed tomographic angiography and positron emission tomography with 18F-fluorodeoxyglucose and the need for their use in different cases is determined. A block diagram of an intelligent diagnostic system for infective endocarditis has been developed. To process the obtained images in order to diagnose and determine the geometric dimensions, shapes, quantity, location, characteristics of infective endocarditis, it is proposed to use convolutional neural networks that allow solving the problem of image segmentation.
Determination of Marketing Parameters for Building a Demand Forecasting Model using Neural Networks
(National Aviation University, 2023-12-27) Sineglazov, Victor; Синєглазов, Віктор Михайлович; Novikov, Mikhaylo; Новіков, Mихайло Сергійович
This article is devoted to finding marketing parameters for building a demand forecasting model using neural networks using real data. The work deals with the problem of modeling product demand on the market in marketing using artificial intelligence and machine learning methods. The main features of existing approaches to building models of products on the market, their advantages and disadvantages are shown. The need for their improvement has been identified. A new methodology for solving the problem is presented. The model's demonstrated ability to predict consumer demand based on a variety of marketing parameters helps businesses plan inventory, production, and personnel more effectively and can lead to significant cost savings and improved efficiency.
Intelligent Medical Image Processing System Using Zero-shot Learning
(State University "Kyiv Aviation Institute", 2024) Sineglazov Victor; Reshetnyk Oleksii
The work is devoted to the intelligent diagnosis of malignant skin tumors. The classification of malignant skin tumors is presented. The greatest attention was paid to skin melanoma. The modern signs of melanoma were analyzed: Asymmetry, Boundary, Color, and Diameter, and additionally for nodular melanoma: Elevated, Firm, and Growing. A review of works on using artificial intelligence to diagnose malignant skin tumors was performed. A methodology for the intelligent diagnosis of malignant skin tumors was proposed, which is based on the use of preprocessing of dermatoscopic images and solving the segmentation problem based on the use of a hybrid approach, which includes the use of a Segment Anything model based on the combination of the Zero-shot learning model, which consists of an image encoder, prompt encoder, lightweight mask decoder, with YOLOv11. ISIC 2018 was used as the dataset. Роботу присвячено інтелектуальній діагностиці злоякісних пухлин шкіри. Представлено класифікацію злоякісних пухлин шкіри. Найбільшу увагу було приділено меланомі шкіри. Проаналізовано сучасні ознаки меланоми: Asymmetry, Boundary, Color, Diameter та додатково для вузлової меланоми: Elevated, Firm, Growing . Виконано огляд робіт з використання штучного інтелекту у діагностиці злоякісних пухлин шкіри. Запропоновано методологію інтелектуальної діагностики злоякісних пухлин шкіри, яка базується на використанні попередньої обробки дерматоскопічних зображень та розв’язанні задачі сегментації на основі використання гібридного підходу, який включає застосування Segment Anything model на основі об’єднання моделі Zero-shot learning, яка складається з image encoder, prompt encoder, lightweight mask decoder з YOLOv11.В якості датасету було використано ISIC 2018.
Language Model Adaptation for Legal Ukrainian Domain
(National Aviation University, 2024) Sineglazov Victor; Savenko Illia
Language models in recent decades make a huge step towards solving the tasks that previously could be done only by humans. Development of NLP area is different scopes gives an opportunity to solve domain specific tasks and transfer knowledge from learnt data towards the useful inferences based on that. This article provides the NLP model approach in specific legal domain. Additionally, this article explores performance of pre-training small models and its utilization and checks the scores on fine-tuned task of checking sentence similarities via SBERT. According to this articles it is proven that domain-specific pre-trained models can perform better results than generally trained language model. This article also provides the language model that is adopted to the Ukrainian legal domain. В роботі розглянуто способи векторизації текстових властивостей природної мови в контексті задачі інтелектуального аналізу тексту. Проаналізовано найпоширеніші способи статистичного аналізу вилучення ознак та методи з урахуванням контексту. Проведено опис вищезазначених типів обрамлення тексту та їх найпоширеніші реалізації. Виконано їх порівняльний аналіз, який показав зв’язок між типом задачі інтелектуального аналізу тексту та методом, що показує найкращі метрики. Описано та реалізовано топологію нейронної мережі, яка стоїть в основі вирішення задачі та отримання метрик. Порівняльний аналіз проведено за допомогою відносного аналізу часу теорії алгоритмів та метрик класифікації: accuracy, f1-score, precision, recall. Метрики класифікації узято з результатів побудови моделі нейронної мережі з використанням описаних методів обрамлення. В результаті в задачі аналізу тональності тексту найкращим виявився статистичний метод обрамлення на основі n-грамів символьних послідовностей.
Long-term Demand Forecasting: using an Ensemble of Neural Networks to Improve Accuracy
(National Aviation University, 2023-09-29) Sineglazov, Victor; Синєглазов, Віктор Михайлович; Samoshyn, Andrii; Самошин, Андрій Олександрович
This research paper proposes a method of long-term demand forecasting based on an ensemble of neural networks that considers the novelty of the data. A tool for creating the ensemble was developed that uses a bagging technique as well as a modification that allows for the relevance and novelty of the data to be considered when creating training samples for each model in the ensemble. The study examines and compares the developed method with known approaches to long-term demand forecasting. Experimental results have indicated that the proposed approach allows for obtaining more accurate and reliable demand forecasts compared to existing methods. The results emphasize the importance of data in the demand forecasting process and indicate the potential of the proposed method to eventually improve inventory management strategies and product planning.
Modification of Semi-supervised Algorithm Based on Gaussian Random Fields and Harmonic Functions
(National Aviation University, 2023-06-27) Sineglazov, V. M.; Синєглазов, Віктор Михайлович; Chumachenko, O. I.; Чумаченко, Олена Іллівна; Lesohorskyi, K. S.; Лесогорський, Кирило Сергійович
In this paper we propose an improvement for a semi-supervised learning algorithm based on Gaussian random fields and harmonic functions. Semi-supervised learning based on Gaussian random fields and harmonic functions is a graph-based semi-supervised learning method that uses data point similarity to connect unlabeled data points with labeled data points, thus achieving label propagation. The proposed improvement concerns the way of determining similarity between two points by using a hybrid RBF-kNN kernel. This improvement makes the algorithm more resilient to noise and makes label propagation more locality-aware. The proposed improvement was tested on five synthetic datasets. Results indicate that there is no improvement for datasets with big margin between classes, however in datasets with low margin proposed approach with hybrid kernel outperforms existing algorithms with a simple kernel.
Optimizing Drone Coverage in Agriculture: an Overview and New Approaches
(State University "Kyiv Aviation Institute", 2025) Sineglazov Victor; Koniushenko Roman
This article investigates the problem of trajectory optimization for unmanned aerial vehicles during multispectral imaging of agricultural lands within the framework of precision agriculture concepts. The main problems related to complex field geometry, presence of natural and artificial obstacles, as well as limited battery capacity of drones are considered. A new hybrid route optimization method is proposed that integrates the ant colony optimization algorithm for global planning of zone traversal sequence with the binary gridding method for detailed local replanning within complex areas and obstacle avoidance. A key feature of the method is an adaptive mission recovery mechanism that allows the drone to dynamically return to the charging station, save mission state, and automatically continue operation from the last uncovered area. Simulation and comparative analysis results demonstrate that the developed approach significantly reduces total traveled route length and optimizes mission execution time compared to traditional methods, confirming its effectiveness for increasing autonomy and productivity of agricultural unmanned aerial vehicles. У статті досліджено проблему оптимізації траєкторії руху безпілотних літальних апаратів під час виконання мультиспектральної зйомки сільськогосподарських угідь у рамках концепції точного землеробства. Розглянуто основні проблеми, пов’язані зі складною геометрією полів, наявністю природних та штучних перешкод, а також обмеженою ємністю акумуляторів дронів. Запропоновано новий гібридний метод оптимізації маршруту, який інтегрує алгоритм мурашиних колоній для глобального планування послідовності обходу зон з методом двійкового сіткового поділу для детального локального перепланування всередині складних ділянок та обходу перешкод. Ключовою особливістю методу є адаптивний механізм відновлення місії, що дозволяє дрону динамічно повертатися на зарядну станцію, зберігати стан місії та автоматично продовжувати роботу з останньої непокритої ділянки. Результати моделювання та порівняльного аналізу демонструють, що розроблений підхід значно зменшує загальну довжину пройденого маршруту та оптимізує час виконання місії порівняно з традиційними методами, що підтверджує його ефективність для підвищення автономності та продуктивності аграрних безпілотних літальних апаратів.
Recommender Systems Based on Reinforced Learning
(National Aviation University, 2023-06-27) Sineglazov, V. M.; Синєглазов, Віктор Михайлович; Sheruda, A. V.; Шеруда, Андрій Володимирович
This article is devoted to the problem of building recommender systems based on the use of artificial intelligence methods. The paper analyzes the algorithms of recommender systems. Analyzes the Markov decision-making process in the context of recommender systems. Approaches to the adaptation of reinforcement learning algorithms to the task of recommendations (transition from the task of supervised learning to the task of reinforcement learning) are considered. Reinforcement learning algorithms Deep Deterministic Policy Gradient and Twin Delayed DDPG were implemented with their own environment simulating the user's reaction, and the results were compared. The structure of a recommender system has been developed, in which the recommender agent generates a list of offers for an individual user, using his previous history of ratings. In the system itself, the user has the ability to interact only with the space of recommended films. This can be compared to the main YouTube page, which is a feed with suggestions, but we have a user interacting only with this feed and his reaction to objects in the recommendation space falls into recommender agent, which regulates the parameters of the model in the learning process.
Semi-supervised Learning Based on Graph Stochastic Co-Training
(National Aviation University, 2023-09-29) Sineglazov, Victor; Синєглазов, Віктор Михайлович; Yarovyi, Serhi; Яровий, Сергій Сергійович
This article is devoted to the development of a new approach in semi-supervised machine learning. The goal of this article is to analyze the accuracy of the single-view co-training system, based on the use of a modified graph-based stochastic label propagation algorithm for a multiclass classification problem. Graph transformation of data is preceded by feature decomposition, with three algorithms being compared: Singular Value Decomposition, Truncated Singular Value Decomposition, Iterative Primary Component Analysis, Kernel Primary Component Analysis. To improve the accuracy of the proposed method, additional parameter was included in the label propagation algorithm, allowing for the usage of the algorithm in co-training systems. Further performance increases are achieved via optimization of data modification, which is achieved by applying feature decomposition methods and parallelizing the calculation-heavy processes. As examples of practical use were considered solutions to the problem of multiclass classification for standard datasets of the library sklearn and for the real dataset Traffic Signs Preprocessed. Analyses of the results of the implementation of the proposed approach showed improvements in accuracy and of performance solving the multiclass classification problem.
Semi-supervised Multi-view Ensemble Learning with Consensus
(National Aviation University, 2024) Sineglazov Victor; Lesohorskyi Kyrylo
This paper is devoted to enchasing existing multi-view semi-supervised ensemble learning algorithms by introducing a cross-view consensus. A detailed overview of three state-of-the-art methods is given, with relevant steps of the training highlighted. A problem statement is formed to introduce both semi-supervised framework and consider the semi-supervised learning in the context of optimization problem. A novel multi-view semi-supervised ensemble learning algorithm called multi-view semi-supervised cross consensus (MSSXC) is introduced. The algorithm is tested against 5 synthetic datasets designed for semi-supervised learning challenges. The results indicate improvement in the average accuracy of up to 10% in comparison to existing methods, especially in low-volume, high density scenarios. Статтю присвячено вдосконаленню існуючих алгоритмів напівкерованого ансамблевого багатовидового навчання шляхом введення консенсусу між видами. Подано детальний огляд трьох найсучасніших методів із виділенням відповідних етапів навчання. Формується постановка задачі, щоб представити як напівкеровану структуру, так і розглянути напівкероване навчання в контексті проблеми оптимізації. Представлено новий багатовидовий напівкерований ансамблевий алгоритм навчання під назвою багатовидовий напівкерований перехресний консенсус (MSSXC). Алгоритм перевірено на п’яти синтетичних наборах даних, призначених для напівкерованого навчання. Результати вказують на підвищення середньої точності до 10% порівняно з існуючими методами, особливо в сценаріях з малим обсягом і високою щільністю.
Semi-supervised Support Vector Machine
(National Aviation University, 2023-03-27) Sineglazov, V. M.; Синєглазов, Віктор Михайлович; Samoshyn, A. O.; Самошин, Андрій Олександрович
The article considers a new approach to constructing a support vector machine with semi-supervised learning for solving a classification problem. It is assumed that the distributions of the classes may overlap. The cost function has been modified by adding elements of a penalty to it for labels not in their class. The penalty is represented as a linear function of the distance between the label and the class boundary. To overcome the problem of multicriteria, a global optimization method known as continuation is proposed. For a combination of predictions, it is suggested to use the voting method of models with different kernels. The Optuna framework was chosen as the tool for configuring hyperparameters. The following were considered as training samples: type_dataset, banana, banana_inverse, c_circles, two_moons_classic, two_moons_tight, two_moons_wide.

Browse

Browsing Кафедра авіоніки та систем управління by Subject "004.855.5(045)"