F3 Комп'ютерні науки

Permanent URI for this collection

https://ekmair.ukma.edu.ua/handle/123456789/18301
Освітня програма: "Комп'ютерні науки"

Browse

Browsing F3 Комп'ютерні науки by Subject "bachelor`s thesis"

Now showing 1 - 4 of 4
  • Thumbnail Image
    Item
    Comparative Analysis of Development Environments for UAV Software Development
    (2025) Budilova, Sofiia; Kurochkin, Andrew
    Unmanned aerial vehicles are in demand nowadays due to their ability to perform various tasks, both military and civilian, without the involvement of humans. The UAV simulators (UAV algorithm development environments) have a great value at the present day since they present a way to test all the new UAV algorithms that are spreading more and more finding real-world applications across various disciplines. A plethora of simulators already exist. All of them have their own advantages and disadvantages. This presents difficulties for developers to opt for the most suitable one to meet their requirements. This paper reviews the most popular UAV simulators. It also provides statistics regarding the amount of papers in general and throughout the years related to simulators and also to flight control software (e.g., ArduPilot). This provides the opportunity to observe tendencies in UAV simulation technologies. Subsequently, research was conducted to analyze scenes (usually called worlds) of the most popular and robust simulator, namely Gazebo, that are available open-source online. тA list of them is provided in this thesis. Eventually, a research gap was found, namely a shortage of Gazebo worlds containing moving objects. A new world with a moving car was created and might be used, for example, for the testing of the UAV object-tracking algorithms.
  • Thumbnail Image
    Item
    Normalization as a Key Enabler for Transferable Machine Learning in Multi-Temporal Cross-Dataset Satellite Imagery: Evidence in Cloud Detection
    (2025) Полякова, Любов; Кундік, Кирило
    In this thesis, we explore the use of normalization and standardization to improve the transferability of deep learning models for cloud detection from multi-temporal satellite imagery. Specifically, we evaluate whether applying normalization techniques during preprocessing can reduce the necessity of model fine-tuning when encountering temporally shifted and externally sourced satellite images.
  • Thumbnail Image
    Item
    Object detection model pruning with application to human recognition in UAV footage
    (2025) Безбородов, Владислав; Кузьменко, Дмитро
    This thesis explores the application of the model optimization techniques in object detection field with a focus on human recognition from UAV footage. Limitations of resource constrained devices and deployment of accurate yet lightweight models is a challenging task. To address this, we examine three core optimization approaches: quantization, pruning, and knowledge distillation. Each method is investigated and applied in the context of YOLOv8-based detectors. Through experimental evaluation and comparative analysis with models trained from scratch, we demonstrate these techniques can significantly reduce model size and inference latency while preserving favorable performance.
  • Thumbnail Image
    Item
    Retrieval Augmented Generation for Ukrainian Government Services: A Comparative Evaluation of the Approaches
    (2025) Маринич, Антон; Курочкін, Андрій
    Retrieval Augmented Generation or RAG is a method that is used to improve the quality of retrieval for LLMs, to avoid hallucinations and be aware of all the changes in the data. This approach integrates LLMs with external data source by building a vector index. This thesis presents a comprehensive study on how different RAG approaches perform in Ukrainian Governmental Services domain. I establish a non-RAG baseline using GPT-4.1-mini model and iteratively perform tests on different configurations of RAG approaches. I have also created a dataset with 500 open questions about Ukrainian Governmental Services using GPT-o4-minihigh model. My best results comparing to the baseline are 13.25% improvement in LLM Judge Score using CRAG with Hypothetical Document Embedding and Reranking and 10% improvement on Factual Correctness using CRAG with Reranking.