Institutional Repository of Vilnius University
Corrosion detection on steel panels using semantic segmentation models /
Abstract
dc:descriptionThe effectiveness of several semantic segmentation networks, including U-Net, FPN, PSPNet, and LinkNet, is examined in this study for the purpose of corrosion detec- tion on various steel panels that have been affected by various chemicals that cause corrosion to emerge. Since these types of images lack publically accessible ground truth datasets, an image prepossessing algorithm was developed and is now employed both fully automatically and semi-manually. Both approaches are compared in or- der to determine the significance of image masking accuracy and human involvement throughout the process. The aim is to examine the performance of neural networks by comparing their performance metrics not only between the models used in this study, but also with those from earlier studies that used fully automated corrosion detection algorithms or various deep learning architectures. The findings indicate that manually creating ground truth datasets has a significant impact on model accuracy metrics, and that only models trained with these types of datasets may be employed successfully in real-world applications. Additionally, compared to various corrosion detection techniques used in other reviewed researches, all architectures tested in this thesis worked well and demonstrated superior results in majority of the used indicators.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Micikevičius, Mantas,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD157854085&prefLang=en_US