also reported higher classification accuracy from SVMs in comparison to the other
two classifiers, with a difference in overall accuracy of ∼2–5%. Karimi et al. (2006)
evaluated the usefulness of SVMs and ANN for hyperspectral imagery classification
over an agricultural area using data from an airborne hyperspectral sensing system
flown over a region in Canada. Authors also found the SVMs to outperform the ANN
classifier by approximately 15% and 0.114 in overall accuracy and kappa coefficient,
respectively. Pal (2006) examined the combined use of SVMs with hyperspectral
imagery from the DAIS airborne sensor for a test region in Spain and reported overall
classification accuracy for SVMs higher that 91%. Koetz et al. (2008) examined the
combined use of hyperspectral and lidar data with SVMs for fuel types mapping for a
region in France. Other authors reported an overall accuracy and kappa coefficient of
69.15% and 0.645, respectively, when the SVMs was applied only with the hyperspectral imagery, which increased by 6.3% and 0.115, respectively, when the two data
sets were combined. More recently, Petropoulos et al. (2012a) evaluated the combined use of Hyperion imagery with SVMs and ANN classifiers for a heterogeneous
region in Greece. Their results showed a close classification accuracy between the two
classifiers (higher than 85% in both cases), with the SVMs somehow outperforming
the ANN by 3.31% overall accuracy and by 0.038 kappa coefficient (Figure 15.5). In
another study, the same authors also compared SVMs and object-based classification
for another region in Greece representative of typical Mediterranean conditions.
Findings from their work showed that both classifiers were able to produce comparatively accurate land cover maps of the studied area. Overall accuracy and kappa for
object-based classification were 81% and 0.779, respectively, whereas for SVMs were
76% and 0.719 respectively.
FIGURE 15.5 Hyperion pixel-based classification using SVM classifier (left) and ANNs
(right). Adopted from Petropoulos et al. (2012a).
HYPERSPECTRAL REMOTE SENSING IN LAND COVER EXTRACTION
309
two classifiers, with a difference in overall accuracy of ∼2–5%. Karimi et al. (2006)
evaluated the usefulness of SVMs and ANN for hyperspectral imagery classification
over an agricultural area using data from an airborne hyperspectral sensing system
flown over a region in Canada. Authors also found the SVMs to outperform the ANN
classifier by approximately 15% and 0.114 in overall accuracy and kappa coefficient,
respectively. Pal (2006) examined the combined use of SVMs with hyperspectral
imagery from the DAIS airborne sensor for a test region in Spain and reported overall
classification accuracy for SVMs higher that 91%. Koetz et al. (2008) examined the
combined use of hyperspectral and lidar data with SVMs for fuel types mapping for a
region in France. Other authors reported an overall accuracy and kappa coefficient of
69.15% and 0.645, respectively, when the SVMs was applied only with the hyperspectral imagery, which increased by 6.3% and 0.115, respectively, when the two data
sets were combined. More recently, Petropoulos et al. (2012a) evaluated the combined use of Hyperion imagery with SVMs and ANN classifiers for a heterogeneous
region in Greece. Their results showed a close classification accuracy between the two
classifiers (higher than 85% in both cases), with the SVMs somehow outperforming
the ANN by 3.31% overall accuracy and by 0.038 kappa coefficient (Figure 15.5). In
another study, the same authors also compared SVMs and object-based classification
for another region in Greece representative of typical Mediterranean conditions.
Findings from their work showed that both classifiers were able to produce comparatively accurate land cover maps of the studied area. Overall accuracy and kappa for
object-based classification were 81% and 0.779, respectively, whereas for SVMs were
76% and 0.719 respectively.
FIGURE 15.5 Hyperion pixel-based classification using SVM classifier (left) and ANNs
(right). Adopted from Petropoulos et al. (2012a).
HYPERSPECTRAL REMOTE SENSING IN LAND COVER EXTRACTION
309