Institutional Repository of Vilnius University
Durpių laukų topografinių vaizdų segmentavimas naudojant giliojo mokinimosi modelius /
Abstract
dc:descriptionThis thesis explores the segmentation of peat field topographic images using deep learning methods. Image segmentation is a core task in computer vision, enabling the automatic identi‑ fication and delineation of meaningful objects in images. The study evaluates both a pretrained zero‑shot model – the Segment Anything Model (SAM) – and several custom‑trained deep seg‑ mentation models, including U‑Net++, DeepLabV3+, FPN, and PSPNet. Experiments were conducted on real‑world topographic images in GeoTIFF format, repre‑ senting various peat field structures. The data were tiled, segmented using selected models, and processed using Python and GDAL tools. The evaluation focused on segmentation accuracy (IoU metric), processing time, and applicability in industrial workflows. Results show that the SAM model performs well on large, well‑defined objects but requires significant computational resour‑ ces. In contrast, the trained deep models achieved high accuracy (up to 0.89 IoU) in considerably less time, making them more suitable for integration into automated industrial systems. The work resulted in a functional segmentation pipeline, enabling the practical application of deep learning‑based models for peat field analysis.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jankūnas, Mantas,
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:ELABAETD239030686&prefLang=en_US