Effect of hyperspectral image denoising with PCA and total variation on tree species mapping using Apex Data

Frieke Vancoillie, Wenzhi Liao, Flore Devriendt, Sidharta Gautama, Robert De Wulf, Kris Vandekerkhove, Loannis Gitas (Editor), Giorgios Mallinis (Editor), Petros Patias (Editor), Dimitris Stathakis (Editor), Georgios Zalidis (Editor)

Research output: Contribution to journalArticle

1 Downloads (Pure)


In this paper, the impact of image denoising on feature selection and tree species mapping accuracy is assessed. We apply a novel algorithm for hyperspectral (HS) image denoising using principal component analysis (PCA) and total variation (TV). The method is embedded in an object‐based classification framework and tested for complex forests with closed canopies and scarce reference data. Results show that, under the given conditions, HS image denoising is beneficial yielding stable mapping results with acceptable accuracy levels. Denoising also affected feature selection processing time with a time gain of 41.6%.
Original languageEnglish
Pages (from-to)281-286
Number of pages6
JournalSouth-Eastern European Journal of Earth Observation and Geomatics
Issue number25
Publication statusPublished - 24 May 2014
Event5th International conference on Geographic Object-Based Image Analysis (GEOBIA 2014) - Thessaloniki, Greece
Duration: 21 May 201424 May 2014


  • denoising
  • broadleaved forest
  • object‐based tree species classification
  • hyperspectral data

Cite this