Institutional Repository of Vilnius University
Hiperspektrinių vaizdų spektrinio išmaišymo metodų lyginamoji analizė /
Abstract
dc:descriptionHyperspectral imaging captures hundreds of spectral bands per pixel, where spectral signatures of several materials usually mix. Spectral unmixing decomposes mixed pixels into constituent materials (endmembers) and their proportions (abundances) is therefore main objective for hyperspectral image analysis. This paper presents the theoretical foundations of spectral mixing models and unmixing method types followed by a comparative experimental analysis of five methods: geometric approaches (VCA-FCLSU, SISAL-FCLSU), sparse regression (SUnSAL), non-negative matrix factorization based method (MVC-NMF), and deep learning method (MiSiCNet). Experiments conducted on real and synthetic datasets evaluate method accuracy, robustness to noise, dependence on pure pixel assumption and computational efficiency. Results show that no single best method exists, SISAL-FCLSU excels when pure pixels are absent, MiSiCNet provides best noise robustness, while geometric methods offer 10–100× faster computation. At the end of the paper practical recommendations for method selection based on data characteristics are provided.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Žiliukas, Augustinas,
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:ELABAETD277690576&prefLang=en_US