DocumentCode :
2334569
Title :
Metric learning for hyperspectral image segmentation
Author :
Bue, Brian D. ; Thompson, David R. ; Gilmore, Martha S. ; Castaño, Rebecca
Author_Institution :
Electr. & Comput. Eng., Rice Univ., Houston, TX, USA
fYear :
2011
fDate :
6-9 June 2011
Firstpage :
1
Lastpage :
4
Abstract :
We present a metric learning approach to improve the performance of unsupervised hyperspectral image segmentation. Unsupervised spatial segmentation can assist both user visualization and automatic recognition of surface features. Analysts can use spatially-continuous segments to decrease noise levels and/or localize feature boundaries. However, existing segmentation methods use task-agnostic measures of similarity. Here we learn task-specific similarity measures from training data, improving segment fidelity to classes of interest. Multiclass Linear Discriminant Analysis produces a linear transform that optimally separates a labeled set of training classes. This defines a distance metric that generalizes to new scenes, enabling graph-based segmentations that emphasizes key spectral features. We describe tests based on data from the Compact Reconnaissance Imaging Spectrometer (CRISM) in which learned metrics improve segment homogeneity with respect to mineralogical classes.
Keywords :
image segmentation; learning systems; compact reconnaissance imaging spectrometer; hyperspectral image segmentation; metric learning; multiclass linear discriminant analysis; task specific similarity measures; unsupervised spatial segmentation; Euclidean distance; Hyperspectral imaging; Image segmentation; Impurities; Materials; Training; CRISM; Metric Learning; Segmentation;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), 2011 3rd Workshop on
Conference_Location :
Lisbon
ISSN :
2158-6268
Print_ISBN :
978-1-4577-2202-8
Type :
conf
DOI :
10.1109/WHISPERS.2011.6080873
Filename :
6080873
Link To Document :
بازگشت