Institutional Repository of Vilnius University
Skiriamąją gebą gerinančio modelio taikymas kaip duomenų augmentacijos metodas laivų aptikimo palydovinėse nuotraukose uždavinyje /
Abstract
dc:descriptionThis paper presents the application of superresolution convolutional neural networks as a data augmentation method to improve the performance of ship detection models. It highlights the current shortcomings of ship detection systems, compares ship detection datasets, and reviews the topics of data augmentation and object detection from satellite imagery. It also describes how the data was collected for the experiments, how the superresolution model was applied, how the ship detection models were trained and how the results were evaluated. The main focus of the study was how different amounts of superresolved images in the ship detection training dataset affect the performance of ship detection. In order to prove this concept, an application was developed that uses superresolution prior to ship detection. Experiments show that the application of the superresolution model as a data augmentation method for training ship detection models improves their performance and makes them more robust to changes in image quality. Also an optimal amount of superresolved images in the ship detection training dataset was found for each superresolution scaling factor.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Gaidys, Kristupas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210645516&prefLang=en_US