Potential of hyperspectral remote sensing for analyzing the urban environment - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 51

Field spectra (solid line) in comparison with
hyperspectral image spectra (HyMap): dotted line =
Dresden 1999, dashed line = Dresden 2000, dash-
dots line = Potsdam 1999

Feature type Feature

function a

Wavelength range
[nm]

(a) Polyethylene (391 image samples)

0.8

Brightness
Increase
Decrease
Increase
Absorption
Decrease
Absorption
Increase
Decrease

1,2
3
3
3
4
3
4
3
3

445 − 2448
486 − 880
1130 − 1202
1202 − 1259
1130 − 1259
1633 − 1721
1633 − 1804
1721 − 1804
2151 − 2305

0.6

0.4

0.2

500 1000 1500

Wavelength [nm]

2000

(b) Red loose chippings (1078 image samples)

0.5

Brightness
Absorption
Absorption
Absorption
Absorption

1, 2
5, 6
5, 6
5, 6
4

445 − 2448
445 − 607
622 − 758
773 − 1085
2134 − 2272

0.4

0.3

0.2

0.1

500 1000 1500 2000

Wavelength [nm]

(c) Asphalt (253 image samples)

0.20
0.18
0.16
0.14
0.12
0.10
0.08
0.06

Brightness
Constant
Increase

1, 2
9, 10, 11

445 − 2448
1169 − 1769

500 1000 1500 2000

Wavelength [nm]

a (1) mean, (2) standard deviation, (3) ratio, (4) area, (5) absorption depth, (6) absorption position, (7) reflectance
height, (8) reflectance position and (9) offset, (10) gain and (11) RMS of a regression line

FIGURE 4.2 Description of interactively identified spectral features for selected urban surface materials. The small step at
about 1000 nm in the reflectance signature of asphalt results from a spectral jump between the VNIR and SWIR I detectors
which is typical for the ASD field spectrometers.

noise fraction (MNF) transformation (Green et al ., 1988), and
the singular value decomposition (SVD) (Golub and van Loan,
1996: chapter 12.4) are widely used. The PCA seeks for principal
components that explain most of the variance of the dataset.
The MNF transformation optimizes the SNR whereas the SVD
results in the projection that leads to the best representation of
the data in the maximum power sense. All of them reduce the
feature space to a significantly smaller number of bands allowing

the determination of the inherent dimensionality of the data as a
prerequisite for optimal endmember detection.

4.3.1.2 Endmember detection

Endmembers are represented by distinct spectral signatures of
surfacematerials. The correct determinationof such endmembers