mi 
   
bands over 
ISS, Bands 
ked objects 
1), channel 
  
m) 
f the solar 
uilding ‘À 
lawn (Fig- 
from areas 
and a con- 
Images in thermal infrared The emitted radiation in the 
thermal infrared part of the spectrum is imaged by sensors as 
'heat' or radiant temperature of the object sensed. The mea- 
sured radiant temperature depends on the objects’ kinetic or 
‘true’ temperature and emissivity Although the skin depth 
from which a thermal image is produced is very thin, subsur- 
face conditions affect surface temperature (Jensen, 1983) 
Because of the differential heating and cooling of objects 
(e g.. water, building, trees, grass, soil), periods of ‘crossover’ 
can occur twice a day. That is, the object's heat emission 
will coincide with that of the background. This fact must be 
considered by the interpreter For example, water bodies are 
cooler than soils and rocks during the day. but owing to the 
higher thermal inertia of the water the temperature reverses 
at night. Fortunately, there are a few objects that do not ex- 
hibit this complex temporal changes in relative temperature. 
Paved roads and parking lots are relatively warm during day 
and night. The pavement surface heats up during daytime 
and the temperature is higher relative to the surroundings 
During the night, heat dissipates relatively slowly. The im- 
age in Figure 7? shows many typical features of daytime im- 
ages. The coolest objects are the water bodies ('2'). Trees 
and shrubs appear cooler than their surroundings ( 3'), while 
driveways, roads (1') and houses ('A', 'B', 'C') are warmer. 
In contrast to most natural surfaces, man-made materials 
have irregular emissivities. Some materials found in urban 
scenes may have the same kinematic temperatures but a com- 
pletely different radiant emittance. Therefore they appear 
very differently in thermal images. For example, unpainted 
metal roofs have a very low emissivity (0.1-0.2) causing ex- 
tremely low gray values in the thermal images (Jensen. 1983). 
Normalized Difference Vegetation Index (NDVI) There 
is no theoretical solution yet for the rigorous inversion of the 
spectral measurements into different properties of the sur- 
face, such as moisture content and vegetation properties 
Vegetation indices are probably the most popular empirical 
tools to analyze remote sensing data. Due to the presence of 
chlorophyll and other absorbing pigments in the leaves, live 
vegetation absorbs the major part of the solar radiation in 
the visible band Consequently. it appears dark in panchro- 
matic photographs There is a sharp increase in reflectance 
in the NIR band. caused by the refractive index discontinuity 
on the boundary of the leaf cells The Normalized Difference 
Vegetation Index, NDVI=(NIR-red)/(NIR+red) depicts the 
relative magnitude of this increase and serves as a vegetation 
indicator The higher the NDVI value, the higher the proba- 
bility that a pixel is at least partially covered by vegetation 
Figure ?? shows the NDVI for our demonstration site The 
color scheme is inverted, that is, light tones are associated 
with low NDVI indicating areas without vegetation. A simple 
segmentation of the image separates man-made objects. For 
example, 48 out of 50 buildings are detected in this fashion 
Classification Here we briefly describe results obtained from 
unsupervised classification. Detailed results of a more rigor- 
ous classification approach are reported in a companion pa- 
per (Merényi and Csathé, 1998) The well known ISODATA 
method (Nadler. and Smith. 1993) is able to discriminate 
most land cover types (Figure 7?) The vegetation is clas: 
sified into two classes, woody-vegetation (trees and shrubs) 
and non-woody vegetation (eg. grass and lawn) Water 
(27) and shadow pixels (see narrow dark areas north of the 
houses) are classified into the same class, since both are dark 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 339 
E e NN kr u. 3 
Vir IP, = 4 
t IM 
  
Figure 4: Normalized Vegetation Index (NDVI) from the mul- 
tispectral data. Low values (white) are indicating areas with- 
out vegetation, such as open water. roads, driveways and 
roofs The NDVI is medium to high (gray to black) on all 
vegetated areas The highest values (black) are associated 
with mature lawn around the houses 
( un 0 vous fed 
opt dus A th 
NDVI (NIR., ^ Narn rod PS ET | 
throughout the spectra Two different types of man-made 
material have been distinguished (dark material as in. C, 
and bright material as in 'A' and B) Areas of bare soil have 
also been classified into the last two categories (e g., around 
'D) Since there is a noticeable difference between the spec 
tra of the bare soil and that of the man-made objects there is 
a possibility that other, non-statistical methods can provide 
better discrimination between them 
Recommendations As mentioned earlier, pixel-based op- 
erations are dominating in the remote sensing processing 
We argue that other. non pixel based methods should be in 
m