Semantic segmentation of marine biofouling images with attention U-Net

Karlatiras, Ioannis; Καρλατήρας, Ιωάννης

dc.contributor.author	Karlatiras, Ioannis	en
dc.contributor.author	Καρλατήρας, Ιωάννης	el
dc.date.accessioned	2025-09-19T10:43:23Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/62468
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.30164
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc/3.0/gr/	*
dc.subject	Βιορύπανση	el
dc.subject	Μηχανική Μάθηση	el
dc.subject	Μηχανική Όραση	el
dc.subject	Σημασιολογική Κατάτμηση Εικόνων	el
dc.subject	Συνελικτικά Νευρωνικά Δίκτυα	el
dc.subject	Marine Biofouling	en
dc.subject	Machine Learning	en
dc.subject	Computer Vision	en
dc.subject	Semantic Segmentation	en
dc.subject	Convolutional Neural Networks	en
dc.title	Semantic segmentation of marine biofouling images with attention U-Net	en
heal.type	bachelorThesis
heal.classification	Neural Networks	en
heal.classification	Marine Biofouling	en
heal.dateAvailable	2026-09-18T21:00:00Z
heal.language	en
heal.access	embargo
heal.recordProvider	ntua	el
heal.publicationDate	2025-02-27
heal.abstract	Marine biofouling presents significant challenges in the maritime industry, including increased drag, fuel consumption, and maintenance costs. Traditional inspection and mitigation methods are labour-intensive and time-consuming, underscoring the need for automated approaches of biofouling detection and analysis. Various machine learning architectures, particularly convolutional neural networks (CNNs), have been utilised for this purpose. This dissertation aims to bridge an existing literature gap by introducing an enhanced Attention U-Net architecture specifically optimised for the semantic segmentation of marine biofouling in operational conditions. The proposed model incorporates additive soft attention gates within the skip connections of the traditional U-Net framework. This mechanism selectively emphasises relevant features by dynamically adjusting the importance of skip connection inputs, reducing noise and improving segmentation accuracy. The model was trained and tested on an annotated dataset of 352 in-water biofouling images collected from multiple ship hull surveys, provided by diving companies, classification societies and NTUA’s archive. It contains imagery captured under various environmental conditions, which enables better model generalisation. It achieves notable performance on both Dice coefficient and intersection-over-union (IoU) scores, suggesting an advancement in classification and localisation capabilities. The results have promising implications for deployment in automated inspection systems, potentially enhancing the efficiency of hull and offshore structure inspections by reducing manual effort and improving detection accuracy. This study provides a scalable solution to biofouling monitoring, filling a critical gap in the field and contributing to the growing need for autonomous biofouling management systems in the maritime industry.	en
heal.advisorName	Papalambrou, George	en
heal.committeeMemberName	Samuelides, Manolis	en
heal.committeeMemberName	Papadopoulos, Christos	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ναυπηγών Μηχανολόγων Μηχανικών. Τομέας Ναυτικής Μηχανολογίας	el
heal.academicPublisherID	ntua
heal.numberOfPages	101 σ.	el
heal.fullTextAvailability	false