Multisource remote sensing data classification based on consensus and pruning

Jon Atli Benediktsson*, Johannes R. Sveinsson

*Corresponding author for this work

Research output: Contribution to journalLetterpeer-review

44 Citations (Scopus)


Multisource classification methods based on neural networks, statistical modeling, genetic algorithms, and fuzzy methods are considered. For most of these methods, the individual data sources are at first treated separately and classified by either statistical or neural methods. Then, several decision fusion schemes are applied to combine information from the individual data sources. These schemes include weighted consensus theory where the weights of the individual data sources control the influence of the sources in the combined classification. Using all the data sources individually in consensus-theoretic classification can lead to a redundancy in the classification process. Therefore, a special focus in this letter is on neural networks based on pruning and regularization for combination and classification. The considered methods are applied in classification of a multisource dataset.

Original languageEnglish
Pages (from-to)932-936
Number of pages5
JournalIEEE Transactions on Geoscience and Remote Sensing
Issue number4 PART II
Publication statusPublished - Apr 2003

Bibliographical note

Funding Information:
Manuscript received September 20, 2002; revised March 13, 2003. This research was supported in part by the Icelandic Research Council and the Research Fund of the University of Iceland. The authors are with the Department of Electrical and Computer Engineering, University of Iceland, 107 Reykjavik, Iceland. Digital Object Identifier 10.1109/TGRS.2003.812000 Fig. 1. Schematic of a linear opinion pool. x(k) = [x .. . ;x ] is an input data vector; N is the number of data sources, ! is an information class; p(! jx ) is a source-specific posterior probability; and is a source-specific weight that controls the relative influence of data source i. This type of a sum is computed for all classes, and the maximum is used for classification [1], [3].

Other keywords

  • Consensus theory
  • Decision fusion
  • Multiple data sources
  • Neural networks
  • Pruning
  • Remote sensing


Dive into the research topics of 'Multisource remote sensing data classification based on consensus and pruning'. Together they form a unique fingerprint.

Cite this