Nearest feature line embedding approach to hyperspectral image classification

Yang Lang Chang, Jin Nan Liu, Chin Chuan Han, Ying Nong Chen, Tung Ju Hsieh, Bormin Huang

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


In this paper, a nearest feature line (NFL) embedding transformation is proposed for dimension reduction of hyperspectral image (HSI). Eigenspace projection approaches are generally used for feature extraction of HSI in remote sensing image classification. In order to improve the classification accuracy, the feature vectors of high dimensions are reduced to the low dimensionalities by the effective projection transformation. Similarly, the proposed NFL measurement is embedded into the transformation during the discriminant analysis stage instead of the matching stage. The class separability, neighborhood structure preservation, and NFL measurement are also simultaneously considered to find the effective and discriminating transformation in eigenspaces for image classification. The nearest neighbor classifier is used to show the discriminative performance. The proposed NFL embedding transformation is compared with several conventional state-of-the-art algorithms. It was evaluated by the AVIRIS data sets of Northwest Tippecanoe County. Experimental results have demonstrated that NFL embedding method is an effective transformation for dimension reduction in land cover classification of earth remote sensing.

Original languageEnglish
Title of host publicationSatellite Data Compression, Communications, and Processing VIII
StatePublished - 2012
EventSatellite Data Compression, Communications, and Processing VIII - San Diego, CA, United States
Duration: 12 Aug 201213 Aug 2012

Publication series

NameProceedings of SPIE - The International Society for Optical Engineering
ISSN (Print)0277-786X


ConferenceSatellite Data Compression, Communications, and Processing VIII
Country/TerritoryUnited States
CitySan Diego, CA


  • Hyperspectral images
  • Land cover classification
  • Nearest linear line embedding


Dive into the research topics of 'Nearest feature line embedding approach to hyperspectral image classification'. Together they form a unique fingerprint.

Cite this