The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
were sited according to historical records. Taking the ancient
capitals as centers, there are abundant ancient cultural relics
distributed from Huxian county in the west to Weiqu of
Chang’an county in the east and from Qinlin mountain in the
south to Xianyang loess tableland in the north (Zhu 1998). This
area belongs to alluvial and diluvial plains of Wei River and
Feng River, and Quaternary strata are well distributed, of which
the thickness is about 800 meters. Along Feng River banks,
even though long archaeological investigations have been
carried out with traditional archaeological methods since 1931,
little information has been acquired about this area because of
geological and geo-morphological restrictions and longtime
human reclamation and cultivation activities.
According to researches of archeologists from National
Historical Museum of China and Archeological Institute of
Shaanxi Province, loess tablelands of Shenhe tableland and
Xiliu tableland are located at the highest height in the area, with
the altitude of about 440~580m, and the depth of groundwater
table is about 20 ~ 50m, which were suitable for tomb
construction by ancestors. Therefore, the research took the
surrounding area of Renjiazhai village and Xiangji temple
village and Jiali Village in Shenhe loess tableland as an
experimental area. See figure 1.
2. RESEARCH METHODS AND RESULTS
The relationships between surface spectral reflectance and
ground temperature, soil moisture and ingredients of soil are
firstly determined by field measurements. And secondly the
links of vegetation growth, soil ingredients, and soil structure
changes to the characteristics of hyperspectral image are
summarized. Simultaneous field measurements of ground
spectrum, thermal infrared temperature, and soil moisture are
required to determine the thermal radiation and temperature
characteristics of nighttime and daytime hyperspectral images.
Thirdly, endeavoring to analyze and interpret temperature, soil
moisture and ingredients anomalies from hyperspectral images
in the research areas combined with historical and cultural relics
knowledges is key research work, because we can possibly get
lots of anomaly areas worth doing further research. Finally,
field boring exploration and historical analysis that eliminate
false anomalies can verify the anomaly areas.
1.2 Data Acquisition
2.1.1 2.2.1 Visible and Short Wave Infrared Image
Processing
1) Spectral Angle Mapping—SAM
In N dimensional space (N bands), pixel classification is
decided by similarity of image pixel spectrum and the reference.
Regarding the spectral responses of N bands as N dimensional
space vectors, the spectral angle from a pixel to the endmember
represents the matching level, smaller angle means more similar.
See Figure 2, and the calculation formula is (1).
0 - cos
(
\
->
t •
r
- cos 1
->
—»
V
t
r
)
nb
/=1
(nb \' l2 (nb \ 1/2
I'. 2 Z r < 2
V ,•=i ) v »=i /
where 0= spectral angle
nb = total bands used in SAM
t = spectrum of unknown objects (to be classified)
r = reference spectrum (known object)
i = number of bands used
2) Linear Spectral Unmixing
Usually, there is not a sole object in one pixel of remote sensing
image because spatial sampling interval is larger than ground
objects during remote sensing scanning. Linear spectral
unmixing is used to estimate the percentage of components in
one pixel.
OMISII spectrometer developed by Shanghai Institute of
Technical Physics, Chinese Academy of Sciences was applied
for hyperspectral scanning. There are 68 bands from visible,
near infrared, short wave infrared to thermal infrared in OMISII.
Daytime and nighttime scannings were carried out in each
research area, and the ground resolution was 3.6m.
Simultaneous field measurements of ground spectrum, thermal
infrared temperature, and soil moisture were also done in the
scanning area with the flight. Thermal Infrared meter, soil
moisture meter and Field Spectral meter provided the
parameters for calibrations of image and relevant quantitative
analyses.
1.3 Image Processing and Results
General image processing methods, such as color composition,
grey stretching & transforming, are still powerful some times in
extracting information, but for hyperspectral images with much
more bands, the following methods are used to utilize all bands
sufficiently to find information of cultural relics.
3) Matched Filter
Based on user-defined endmembers, each pixel is separated
partly. Matched filter is used to calculate the percentage of
user-defined endmembers of pixel by partial unmixing. The
percentage distributed map of each endmember is determined
by matching spectral curve of pixel in imagery and that of
ground object in known spectral library. This method doesn’t
need all endmembers in image.
4) Spectral Feature Fitting—SFF
Based on characteristics of spectrum absorption, non-classified
ground object spectrum could be matched for classification with
referenced spectrum by using the least squares method.
5) Pseudo Colour Composition and Enhancement