éme A Extract 
4| Meanimage … asphalt pixels 
Pme” with low 
mean value 
    
  
  
Ax 
* 
; iow 
o CWwedstb ^ * 
H 
Medium 
  
Figure 5. Classification using the brightness (mean and standard 
deviation) feature. 
The classification result based on this procedure is shown in figure 
6 for a large area. A closer look to a detail (Figure 7) reveals the 
high quality of the classification. 
M C€onerete 
M Masked Pixela 
  
Figure 6. Classification map (mean and standard deviation 
functions). 
In the classification results (excluding non-road pixels), 42.0 % of 
the pixels are identified as asphalt, 2.3% as concrete, 3.2% as 
gravel and 52.5% of the pixels are unclassified. The number of 
unclassified pixels in the a priori given road layer is still high. It 
should be noted that no post processing was applied to improve 
this numbers. Compared to the SAM result the significant 
reduction of unclassified pixels is obvious. 
  
  
  
  
  
  
  
  
  
  
  
Figure 7. An example of a well classified area. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
3.2 Road surface condition determination 
In addition to the road surface material identification, condition of 
the road surfaces is another critical issue in relation to 
transportation. Recent studies (Gomez, 2002; Herold et al., 2005; 
Noronha et al., 2002) in hyperspectral imagery and spectrometry 
have shown that it is possible to map road surface condition and 
distress using these technologies. In terms of condition 
determination, the investigation is carried out for asphalt. Three 
categories are defined, namely good, intermediate and bad. In 
order to determine the different conditions according to the three 
defined categories, spectral features are used. The spectral features 
tested are brightness using mean function, decrease or increase in 
reflectance using the ratio function and hydrocarbon absorption 
bands. 
Mean is one of the functions which is helpful for asphalt condition 
identification. New asphalt has lower mean and as the condition 
gets worse the mean gets higher. In terms of the ratio function 
(Herold et al, 2005) there are two image ratios which are 
significant in asphalt condition differences. These are situated in 
visible and short wavelength infrared bands. These ranges (Figure 
8) are 490nm - 830nm for visible (VIS2) and 2120 nm- 2340 nm 
for short wavelength infrared (SWIR). Good condition asphalt has 
the lowest value in the VIS2 ratio and highest value in the SWIR 
ratio. The converse is true for bad condition asphalt. Another 
approach for identifying different states of material is to use 
hydrocarbon seeps. These features are typically the maximum 
absorption bands situated around the wavelength of 1730 nm and 
2300 nm (Clutis, 1989). The degree of oily components existing in 
asphalt is one of the factors creating different conditions since this 
characteristic influences the molecular structure and consequently 
the degree of viscosity of the asphalt surface. The higher the oily 
components, the more viscous the road surface and in turn the 
stronger the hydrocarbon absorption bands. Deeper absorption 
bands indicate better condition of the asphalt surface material 
(Figure 9). As the asphalt gets older (condition of the asphalt gets 
worse), the degree of viscosity reduces and it becomes prone to 
cracks (Weng Q., 2008). As a result, the reflectance of the surface 
increases and eventually the hydrocarbon bands become weak and 
approximate a straight line. This means that the condition of 
asphalt is bad. The results of the investigation indicate that the 
mean function is more reliable for identifying different conditions 
of asphalt. The classification result for the whole study area based 
on mean function is shown in figure 10. The classification 
statistics indicate the 23% of the pixels are identified as good 
asphalt, 23% as intermediate and 14% as bad asphalt. The 
remaining pixels are unclassified. The statistics are with respect to 
the area covered by roads only. Figure 11 shows an example of 
part of Steinbeis road which is classified as bad. Spectra of 
corresponding locations from the HyMap data supports the result 
obtained. Additionally, field investigation confirms that the 
condition of the asphalt for the road is bad. 
   
      
  
2000 
Reîleciance[7]} 
1000 
2000 
1500 
RIT 
1000