Developing a system for

computer-assisted detection of
cultural heritage sites

Anke Loska Norwegian Directorate for Cultural Heritage /Riksantikvaren

Rune Solberg Norwegian Computing Center /Norsk Regnesentral
Lars Gustavsen Museum of Cultural History, University of Oslo /Kulturhistorisk museum
Trude Brun Vestfold County Administration/ Vestfold fylkeskommune
Christer Tonning Vestfold County Administration/ Vestfold fylkeskommune
PUBLIKASJONENS NAVN:
Developing a system for computer-assisted detection of cultural
heritage sites

OPPDRAGSGIVER:
Norwegian Directorate for Cultural Heritage/ Riksantikvaren
Norwegian Space Centre/ Norsk Romsenter

UTFØRENDE INSTITUSJON:
Norwegian Directorate for Cultural Heritage/ Riksantikvaren
Norwegian Computing Center/ Norsk Regnesentral
Museum of Cultural History, University of Oslo/ Kulturhistorisk
museum
Vestfold County Administration/Vestfold fylkeskommune
FORFATTERE OG FAGLIG Anke Loska: Norwegian Directorate for Cultural Heritage
ANSVARLIG:
Rune Solberg: Norwegian Computing Center
Lars Gustavsen: Museum of Cultural History, University of Oslo
Trude Brun: Vestfold County Administration
Christer Tonning: Vestfold County Administration

NORSK ROMSENTERS JOP.03072

KONTRAKTSNUMMER
DATO:
Mai 2008
ANTALL SIDER / OPPLAG: 27/75
REG.NR:
08-54
BESTILLING:
[email protected]
www.riksantikvaren.no

Foto: Norsk Regnesentral/Norsk Romsenter/Riksantikvaren

Satelittbildet på framsiden er av et jorde ved Numedalslågen. De fire sirklene på jordet er
tegn på graver. Bildet er tatt av satellitten QuickBird.

2
Executive summary
The increasingly intensive use and modification of the landscape resulting from modern
demands for efficient infrastructure and land use (agricultural production, mining, energy
sources, leisure/tourism facilities, etc.) exerts growing pressure on cultural heritage in the
landscape. In order to match the political intentions of updated and sustainable cultural
heritage management, it is necessary to develop a cost-effective method for locating and
monitoring cultural heritage sites. In recognition of this, a project was started in 2002
with the overall aim of developing a cost-effective method for surveying and monitoring
cultural heritage sites on a regional and national scale.

The first study area was an intensively exploited, agricultural production area in Rygge
Municipality, Østfold County. Later, two geographically separate areas in Vestfold and
Akershus counties were chosen. A crucial aspect was that there have been archaeological
investigations in these areas. Results obtained in the 2002 pilot project indicated the
existence of a correlation between cultural heritage sites and variation in the chemical
elements in the soil. The results demonstrated that high-resolution geo-chemical sampling
appears to be a promising field for the development of cultural heritage indicators.
However, the costs indicated a need for funding which was almost impossible to obtain.

It was then suggested to focus on the development of automated methods, such as pattern
recognition, for detecting and locating cultural heritage sites. The working assumption is
that cultural heritage sites with no visual apparent manifestations above ground may be
detectable in satellite images due to alterations in the spectral signature of the bare soil or
of uniform vegetation growing there (crops).

During the last project years the aim was to develop a software prototype, CultSearcher,
to provide computerised assistance in the analysis of satellite images. In particular, the
software marks possible sites for further inspection by an archaeologist.

The methods currently used in CultSearcher to search for potential cultural heritage sites
are performed in three main steps: Segmentation, feature extraction and classification. In
the first step potentially interesting locations are detected as image segments, in the
second step characteristics of these segments are computed, before the last step
undertakes a classification of the various segments by comparing them to inherent class
descriptors. Before these main steps are performed, the images have to be imported, and
regions/areas of interest must be identified (agricultural fields). After the main steps, the
results need to be checked. In addition, the system contains functionality for interactive
training of the system to recognise and discriminate between the various region types
(classes) of interest and non-interest.

The user site was represented in the project by two organisations, Vestfold County and
the Museum of Cultural History (KHM). The aim of their involvement was twofold:
Firstly, to test whether the software would be capable of detecting actual archaeological
features in satellite imagery; and secondly, it was aimed at testing the suitability and
functionality of the software for use in a cultural heritage management environment. The

3
areas selected consisted of images recorded in Vestfold County and in the vicinity of
Gardermoen airport in Akershus County.

The user representatives concluded that CultSearcher is clearly still a prototype software,
but with significant potential. It can offer archaeologists a better view of what to expect
when prospecting and excavating agrarian landscapes. From visual inspections of satellite
imagery it became evident that archaeological features can in fact be seen from space in
the form of ring-shaped crop marks. An algorithm for detecting these has been developed
and integrated with the system late in the project. It is capable of detecting a significant
number of the ring-shaped patterns. The user interface in CultSearcher and the process
from creating masks and importing satellite images to extracting the final interpretation
of detections is still somewhat complicated. However, it is clear that this prototype has
the potential for further development, and CultSearcher will be of great value for
archaeologists in the field of cultural heritage management when it is fully developed and
made operational.

In the near future (2008) the system will be tested on all the agricultural areas in Vestfold
County. This will give a broader overview of potential obstacles and important
experience for further development of the algorithms. Since there is international interest
in the methodological approach of CultSearcher, further work includes aims of being
active in the national and international research arenas. The Directorate for Cultural
Heritage will work towards covering all the agricultural areas in Norway with satellite
images, resulting in an overview of potential locations of cultural heritage sites
nationwide. The year 2013 is suggested as a possible milestone for this.

4
Contents
1 Introduction................................................................................................................. 6
2 Marks of cultural heritage seen from the skies ........................................................... 8
3 CultSearcher – a prototype system for the detection of cultural heritage sites......... 11
4 Examples of CultSearcher results ............................................................................. 14
4.1 Detection of amorphous soil marks in an Ikonos image................................... 14
4.2 Detection of amorphous soil marks in a Quickbird image ............................... 16
4.3 Detection of ring-structures as crop marks in Ikonos and Quickbird images... 18
5 CultSearcher from an archaeologist’s viewpoint...................................................... 21
5.1 Testing CultSearcher on Quickbird images from the Gardermoen area........... 21
5.1.1 The test area .............................................................................................. 21
5.1.2 The test...................................................................................................... 22
5.1.3 Preliminary observations .......................................................................... 22
5.2 Testing CultSearcher on Ikonos images from Vestfold County ....................... 23
5.2.1 The test area .............................................................................................. 23
5.2.2 Testing various versions of CultSearcher ................................................. 23
5.2.3 Remarks .................................................................................................... 24
6 Plans for the future.................................................................................................... 25
7 References................................................................................................................. 27
Contact information .......................................................................................................... 28

5
1 Introduction
The increasingly intensive use and modification of the landscape resulting from modern
demands for efficient infrastructure and land use (agricultural production, mining, energy
sources, leisure/tourism facilities, etc.) exerts growing pressure on cultural heritage in the
landscape.

In order to match the political intentions of updated and sustainable cultural heritage
management, it is necessary to develop a cost-effective method for locating and
monitoring cultural heritage sites. Given the enormous costs of surveying the areas in
question by traditional fieldwork, alternatives must be sought. The use of modern support
technologies is imperative, if such rapid changes are to be balanced against the
sustainable management of this resource. One possible approach is through the use of
satellite images.

In recognition of this, the Norwegian Directorate for Cultural Heritage (Riksantikvaren,

RA), in collaboration with the Norwegian Computing Center (Norsk Regnesentral, NR),
the Museum of Cultural History (Kulturhistorisk museum, University of Oslo); Vestfold
County Administration (Vestfold fylkeskommune) and the Norwegian Institute for
Cultural Heritage Research (Norsk Institutt for Kulturminneforskning, NIKU1), started in
2002 a project with the overall aim of developing a cost-effective method for surveying
and monitoring cultural heritage sites on a regional and national scale. Additional and
important funding from 2002-2007 was provided by the Norwegian Space Center (Norsk
Romsenter, NRS).

The first study area was an intensively exploited, agricultural production area in Rygge
Municipality, Østfold County. Later, two geographically separate areas in Vestfold and
Akershus counties were chosen. A crucial aspect was that there have been archaeological
investigations in these areas.

Results obtained in the 2002 pilot project indicated the existence of a correlation between
cultural heritage sites and variation in the chemical elements in the soil. A central focus
in the early project years was the manual analysis of satellite images followed by
chemical profiling of sites observed in these images in order to gain experience as to how
cultural heritage sites really manifest themselves in satellite images. The results
demonstrated that high-resolution geo-chemical sampling appears to be a promising field
for the development of cultural heritage indicators. However, the costs involved
demanded a need for funding which was almost impossible to obtain.

It was then suggested to focus on the development of automated methods, such as pattern
recognition, for detecting and locating cultural heritage sites. The working assumption is
that cultural heritage sites with no visual apparent manifestations above ground may be

1
NIKU was involved in the project in 2002-2004.

6
detectable in satellite images due to alterations in the spectral signature of the bare soil or
of uniform vegetation growing there (crops).

During the last project years the aim was to develop a software prototype to provide
computerised assistance in the analysis of satellite images. In particular, the software
marks possible sites for further inspection by an archaeologist. This means that the
archaeologists may focus their efforts on analysing the identified sites. It is important to
bear in mind that the system is designed to detect candidate sites and that no claim is
made that these candidates are true cultural heritage sites. Even human specialists cannot
make such an assertion based on satellite imagery alone. The verification of a potential
site always depends on some kind of field inspection.

Although the costs connected with acquiring and analysing the satellite data will not be
insignificant, and fieldwork will never be replaced entirely by high-technological
methods, it seems plausible that an essentially cheaper, and possibly even qualitatively
better method for the surveying and monitoring of cultural heritage sites can be
developed to target fieldwork to a degree not possible today.

The Norwegian Computing Center has been responsible for developing the automatic
detection, methodology and implementing this into a prototype software system,
CultSearcher. Vestfold County and the Museum of Cultural History were crucial in
selecting areas of interest and in testing the software prototype as a possible tool for
future cultural heritage management. The project was funded by The Directorate for
Cultural Heritage and the Norwegian Space Center, whose representatives acted as the
Steering Committee.

The Norwegian Institute for Cultural Heritage Research was involved in the project
during the early years where geochemical soil analysis was the main focus. The
representative of NIKU continues that approach in collaboration with Vest-Agder County
at chosen areas in southern and western Norway.

7
2 Marks of cultural heritage seen from the skies
While the detection of cultural heritage sites from space is a rather new discipline, such
sites have been observed and detected from the air for about one hundred years. The first
reported cases refer to observations of Stonehenge (1906) and Forum Romanum (1906-
1908) using balloons. The first reported cases using aircraft are concerning ruined towns
and cities in Sinai seen by pilots during World War I. However, the first scientific aerial
archaeology started in the 1920s with the British geographer and archaeologist Osbert
Guy Stanhope Crawford (1886-1957). Due to his pre-war interest in the cartography of
linear earthworks of prehistoric origin and his war experiences as an aerial observer and
photographer, it was Crawford who created this new archaeological discipline.

What the aerial archaeologist typically sees are shadow-marked sites and levelled sites.
Shadow marked sites are sites cut into the soil or rising above it, like castles, ruins,
fortifications (banks and ditches still preserved) or tumuli. The visibility depends on the
preserved height, the colour of the objects, vegetation cover, solar elevation and
observation angle. Levelled sites are traces left at the surface which are only visible under
certain conditions. There are two types: soil-marked sites and crop-marked sites.

Soil-marked sites are typically the remains of ditches, pits, buried walls, etc. A ditch or a
pit will disturb the local soil profile, and refilled material usually has different
characteristics, like density and composition. The refilled material is in most cases not so
compact, and it might contain more humus components, making it looking darker. The
refilled material may also affect the soil texture with a grain-size distribution that differs
from the undisturbed soil (usually larger number of smaller grain sizes). This results in
improved water-storage capacity, so the soil will look darker under certain conditions.

Figure 2.1. Soil marks are due to ditches (left) and roads (right) in aerial photos from
Austria (Aerial Archive, Institute for Prehistory and Protohistory, University of Vienna)

In the case of buried structures like walls etc, remaining compact stones and mortar
cannot store any water and the soil dries easily. Stones and mortar might also be brought
to the surface by ploughing, creating contrast as stones look brighter than the surrounding
soil.

8
Crop marks are an indirect effect of buried archaeological features. Their visibility
depends on the soil, climate and vegetation. So-called positive marks are due to more
water available which makes plants grow higher and ripen later than the surrounding
plants. A colour-tonal contrast may be created because the vegetation stays green for a
longer period and/or that the vegetation is darker green. Crop marks may also be due to
vegetation relief. Plants grow higher, enough to throw a shadow in slanting sunlight. So-
called negative marks appear when plants grow over buried stones (e.g. walls) and run
out of water sooner, ripen earlier and stay shorter. Almost any crop can develop marks, if
conditions are favourable. Cereals react quickly to Soil Moisture Deficit (SMD) and are
growing very close, making the contrasts clearer.

Figure 2.2. Negative crop marks in Austria due to ditches (left) and positive crop marks
in the Czech Republic due to building walls (right) in aerial photos (Aerial Archive,
Institute for Prehistory and Protohistory, University of Vienna)

Geometrical patterns may also appear in agricultural fields as frost marks and snow
marks. Refilled ditches and walls can store heat or cold (having different thermal
capacity). Under the right weather conditions, these might be visible as differential
thawing and freezing of radiation frost (hoar frost) or a thin snow cover. Such marks are
visible just during a few hours time span (typically in the morning).

Even if there are remaining structures rising above the terrain or below the terrain as
ditches and pits, they might be hidden by tall vegetation, in particular forest. If the terrain
can be mapped accurately enough, such archaeological remains might be detectable in
elevation data. These marks are therefore called relief-marked sites.

Sub-surface structures might also be discovered even if they create none of the mark
types described above. Remains of constructions (usually stone constructions) often
create contrast to the surrounding underground material (soil or sand), and can therefore
be detected with sensors emitting and measuring electromagnetic or acoustic signals.

9
Various types of remote sensing sensors, airborne and spaceborne, are useful for
detecting remains or patterns created by cultural heritage sites. Soil- and crop-marked
sites can be measured with high-resolution optical (visible and infrared) sensors. With the
optimal selection of observation wavelengths, high contrast can be obtained (in particular
appearing from reflectance contrasts due to soil moisture or vegetation density). The
spatial resolution of these sensors should be of 1 m or better to be really useful. Frost and
snow marks are also detectable using optical sensors of similar characteristics. Also,
sensitive thermal sensors might be applied.

Figure 2.3. A Roman fortress in England as seen in a lidar image (left) and a town
buried in the sand of a desert in Syria as imaged with a Synthetic Aperture Radar
(English Heritage)

Laser-based sensors, lidars, have got quite a lot of attention recently. Airborne laser
scanning is applied for, e.g., forest mapping. A by-product of this mapping is an accurate
digital terrain model. Relief-marked sites, invisible under tree-cover, may then appear
clearly in such a terrain model.

Radar (in particular Synthetic Aperture Radar) is also of potential interest for remote
sensing of remains of or hidden cultural heritage sites. SAR signals penetrating the
vegetation might interact with the terrain and show relief-marked sites. For dry-ground
conditions, the SAR might also penetrate deeply into the ground. Sub-surface structures
might then appear. Roads and buildings have been found hidden under the sands in
deserts using SAR.

Shadow marks Soil marks Crop marks Relief marks

Everywhere Agricultural areas Agricultural areas Everywhere
All year Spring and autumn Summer All year (dry season
Optical and SAR Optical Optical for SAR)
Lidar and SAR

Figure 2.4. The main types of marks of cultural heritage sites together with an
indication of where they might be found, when they are visible in the annual cycle and
with what type of remote sensing sensors they can be detected

10
3 CultSearcher – a prototype system for the detection
of cultural heritage sites
As explained in the previous section, in a satellite image cultural heritage sites may be
visible as areas with different radiometric and spectral properties compared to their
immediate surroundings. Regardless of their exact cause, cultural heritage sites may show
up as patches, with or without particular shapes, which are darker or brighter than their
surroundings. This is exploited in the methods used to automatically search for such
patches in the satellite images. The CultSearcher prototype is briefly described in the
following. A more detailed description can be found in Amlien et al. 2007.

The methods currently used to search for potential cultural heritage sites are performed in
three main steps: Segmentation, feature extraction and classification. In the first step
potentially interesting locations are detected as image segments, in the second step
characteristics of these segments are computed, before the last step undertakes a
classification of the various segments by comparing them to inherent class descriptors.
Before these main steps are performed, the images have to be imported, and regions/areas
of interest must be identified (agricultural fields). After the main steps, the results need to
be checked. In addition, the system contains functionality for interactive training of the
system to recognise and discriminate between the various region types (classes) of
interest and non-interest.

Segmentation is the process of dividing the areas of the satellite image into different
image regions based on radiometric/spectral and spatial characteristics. The current
segmentation method works on panchromatic images and operates by identifying areas

Figure 3.1. Overview of the main menu structure

11
that are darker or brighter than the surroundings. The method is used in two passes, first
to find dark regions and then to find bright regions.

The characteristics or features extracted in the feature extraction step are typically figures
describing particular geometrical, radiometrical or textural properties of the image
segments extracted in the first step. Examples of such features are complexity, border
quality, aspect ratio, uniformity and contrast. Complexity is calculated as the ratio
between a region’s perimeter and area; border quality is calculated by summing the
gradient information along the border of the region (and normalising by the border
length); aspect ratio is calculated as the ratio of the lengths of the major and minor axes
of an ellipsis adapted to the region; uniformity is the standard deviation of pixel values
within the region; and contrast is the difference in mean grey levels inside the region and
in an area surrounding the region.

Classification is performed on the regions resulting from the segmentation. In this

process, the spatial and radiometric/spectral characteristics of the segmented regions are
taken into consideration to determine whether they are potential cultural heritage sites or
not. During classification, features are extracted from the segmented regions of unknown
class. Based on the statistical class descriptions, a minimum distance classifier is used to
determine the most probable class for each region.

The system is operated through a simple graphical user interface (GUI) that provides the
user with two different methods for running the system; one fully automatic method and
one stepwise method providing some more user control (Figure 3.1). In both approaches
care has been taken to keep the necessary knowledge of technical details to a minimum.
A user with little knowledge of image processing and remote sensing should still be able
to run the system without too much training.

Figure 2. The main modules and the overall system architecture of CultSearcher

12
In addition to letting the user initiate segmentation and classification of the input images,
the interface lets the user generate masks to limit the processing to agricultural fields.
These masks are derived from a digital map giving the boundaries of these fields. The
interface also lets the user load a mask and a satellite image into a common reference
frame.

Finally, the interface provides functionality for training the system (“teaching” the
classification step) and manually deleting classified regions that are deemed to be of no
interest. Figure 2 illustrates the main modules and the overall architecture of the system.
The functionality of each module is briefly described in Table 1.

The user will not need to specify a lot of parameters to the analysis; just a few setup
parameters are defined in the Interactive GUI. For each step in the interactive mode the
user also needs to enter or confirm the file names. Modules, like Edit result, Mark and
Create mask, require more user interaction.

Table 3.1. The main system modules and their functionality

Main GUI The Main GUI is where the user chooses interactive or automatic mode of
processing. In addition, Mark and Training are started directly from the Main GUI.
GUI/Interactive The user controls the processing chain by selecting which processing steps to run.
mode
GUI/Automatic Here the system runs automatically without any user interaction. Note that the steps
mode Create mask and the Edit result are not included here.
Create mask Allows the user to define which areas in the satellite image that will be analysed. It is
available in interactive mode only.
Import image Imports satellite images and masks and prepare them for analysis.
Segmentation Detects potentially interesting locations by performing a segmentation of the satellite
image.
Extract features Extracts image features or characteristics of the detected image segments.
Classification Analyses the extracted features of each detected image segment in order to classify
the site as a potential cultural heritage site or something else
Edit result The user is led through the classified segments one by one, and given the possibility
to delete segments that are not believed to represent cultural heritage sites.
Mark The user is being led through the segments that were identified in the segmentation
module in order to prepare Training. For each segment the user is invited to assign a
class label.
Training An automatic function updating the class descriptions that are used by the
classification module. This is the way to teach the system what potential cultural
heritage sites look like.

13
4 Examples of CultSearcher results
The current version of the CultSearcher prototype system supports computer-assisted
detection of potential cultural heritage sites in agricultural fields as soil and crop marks.
The methods used so far are tailored to detect amorphous image structures (“sites”) as
soil marks and ring structures as crop marks. The system may be extended to cover a far
broader range of structure types as soil and crop marks using the same overall approach.
The current version of the system is restricted to processing images from the Quickbird
and Ikonos satellites. We will in the following provide examples of detecting amorphous
structures as soil marks and ring structures as crop marks in images from both satellites.

We present in the following results from the Oslofjorden region, in particular for areas
surrounding the Lågen River in Vestfold County as well as those surrounding the Rygge
Municipality in Østfold County in Norway. Both areas are rich in known cultural heritage
sites and is also expected to contain a large number of unknown sites. More details can be
found in Grøn et al. 2004 and in Aurdal et al. 2006.

4.1 Detection of amorphous soil marks in an Ikonos image

The development of the CultSearcher methodology started with experiments based on an
Ikonos image data set of the Rygge Municipality acquired in August 2001. The image
data comprised a panchromatic band of 1 m resolution and a set of four multispectral
bands (near-infrared, red, green and blue) of 4 m resolution.

The study area is a typical, intensively

exploited, agricultural production area with a
quite moderate topography in Norwegian
terms. The extent of the study area was more
than 100 square kilometres (Figure 4.1). This
first study concentrated on smaller parts of the
total area. In particular, it was known from
field studies that the areas around the Gipsund
farm, in the north-eastern corner of the total
study area, are rich in cultural heritage sites.
We therefore extracted a sub-image as shown
in Figure 4.2. This sub-image comprises the
central farm area along with the neighbouring
fields. The figure shows as well an Figure 4.1. The Ikonos image over
archaeologist’s indication of possible cultural Østfold county
heritage sites.

14
A three step analysis process was
developed based on this data set:

1. Pre-processing: Agricultural field

masks are derived from land-cover
GIS data in order to restrict further
processing to agricultural fields only.
Each field was then pre-processed in
order to suppress artefacts that could
interfere with the clustering; in
particular plough furrows were
removed. We obtained good results
using Fourier analysis in combination
with mathematical morphology. Figure 4.2. Sub-image showing the Gipsund
farm and surrounding fields. Soil marks that
2. Segmentation: An unsupervised might be of archaeological interest have been
clustering (k-means) was applied to indicated by an archaeologist
each field. This clustered the pixels in
the field according to their spectral
properties.

3. Feature extraction: Each structure

(object) from the segmentation was
then characterised according to shape,
size, contrast, etc. resulting in a
feature vector for each object.

4. Classification: We applied
unsupervised clustering (k-means)
using five classes. An alternative is
supervised classification using, e.g.,
the maximum likelihood approach.
Each object is then classified into
predefined classes, where the
characteristics of the classes are
determined from a set of training
samples. At least one of the classes
should represent potential cultural
heritage site.

Figure 4.3 shows the result of the

classification step, where each class is
indicated by a specific colour.
Figure 4.3. The result of the classification
Based on this initial study, it was step. Each colour indicates a class. The
concluded that a fully automated archaeologist’s annotations are included

15
system would be nearly impossible to develop. The decision as to whether a site is
interesting or not depends on many details. The spectral signature and shape of the site is
only one of several factors that must be taken into account. Knowledge of archaeology
and local history coupled with geography and, obviously, knowledge from field surveys
will often be of primary importance in the final interpretation. Further work in the project
was therefore aimed towards development of a tool for detecting potentially interesting
sites, leaving the final interpretation to the human specialists. Such a tool would greatly
reduce the burden on the human specialist as it would be able to guide the specialist from
site to site in the images. The specialist would then concentrate on the actual
interpretation of the different sites that are detected.

The results from the analysis of the Ikonos image was later compared to the results of a
similar analysis of a Quickbird image acquired in July 2003. It was clear from the
comparison that details visible in one set of data might be more or less invisible in others.
The soil marks seen in the Ikonos image were more or less invisible as crop marks in the
Quickbird image.

4.2 Detection of amorphous soil marks in a Quickbird image

The methods developed in the initial experiments were then tested and further improved
upon a larger data set based on various satellite images acquired in the regions
surrounding Oslofjorden. We present here results from areas surrounding the Lågen River
in Vestfold based on a Quickbird image acquired in 27 April 2005.

Figure 4.4 shows the subsection of the Lågen image discussed here. This region contains
an excavation site (Iron Age grave mounds) as well as several known cultural heritage
site locations within tilled agricultural fields. Map data for the agricultural fields has been
transformed into masks delimiting the processing only to agricultural areas. Some
potential sites are indicated with red arrows. In the segmentation step we seek to detect
these and other potential sites based primarily on their contrast to the local background.
The segmentation was changed here compared to the previous experiments, now using
Niblack’s method (Niblack 1986) for threshold selection. The method is used in two
passes, first to find dark regions and then to find bright regions.

Region features are extracted in the feature extraction step from each region in Figure 4.5,
and the regions are then analysed in the classification step based on these features. A
modified and extended feature set, compared to the previous experiments, was applied
here, and the Maximum Likelihood approach (supervised classification) was applied
instead of unsupervised clustering. The final class of a region was determined by finding
the statistically most likely class given the features.

16
Figure 4.4. Part of the image. Potential Figure 4.5. Result of segmentation applied
cultural heritage sites will typically appear to the image shown in Figure 3
as indicated, either as spots that are darker
or brighter than their surroundings

The result of the classification step is shown in

Figure 4.6 on top of the original image in Figure
4.4. This figure shows only the regions that
belong to either class 1 or 2, that is, the two
classes corresponding to potential cultural
heritage sites.

The current segmentation and classification

methods are designed to be quite inclusive. The
philosophy is that it is worse to loose one real
cultural heritage site than detecting a high
number of false sites. However, if the number
of false positives becomes too high the user will
spend too much time going through the detected
sites. Hence, in future work we will seek to
reduce the number of false positives. There are
numerous ways of achieving this. In
collaboration with archaeologists, we will try to
tune the system better so that the interesting
sites are still detected, while uninteresting sites Figure 4.6. Result of the final classification
to a larger degree are rejected. overlaid on part of the original image

17
4.3 Detection of ring-structures as crop marks in Ikonos and Quickbird
images
Ring structures are of great interest to archaeologists as they may indicate the existence
of remains of burial mounds and other circular structures. Ring-shaped structures may be
the remnants of graves which were originally constructed as burial mounds surrounded
by a ditch. In these ditches combustible material was burned over many centuries. Today,
the mounds themselves have been destroyed by agricultural activity, but the presence of a
thick layer of ashes in the surrounding ditch might still be visible in some regions, see
Figure 4.7.

In the final part of this project, we performed experiments for the detection of ring
structures in agricultural fields. We here present some of these results from crop mark
analysis in Ikonos and Quickbird images acquired 29 July 2003, 13 August 2003 and 30
June 2006. For more details, see Larsen, Trier and Solberg 2008.

The ring structures may appear in

numerous different ways. The circles
vary in size, i.e. radius and width. Some
rings are brighter than their surroundings,
while others are darker. The examples in
Figure 4.7 are all relatively clearly
visible. However, this is not always the
case. Sometimes the remains consist of
circle fragments only, and/or the border
of the ring is more diffuse than what can
be seen here. A few other examples are
shown in Figures 4.9 and 4.10.

The approach of detecting circular

objects is a modification of the approach
for detection of amorphous objects
described in the previous sections.
Figure 4.7. Visible ring structures in a The main idea of the segmentation
satellite image of an agricultural area approach is to search the images for
areas that matches a given ring template. Based on visual observations of a series of 0.6
m resolution Quickbird and 1 m resolution Ikonos panchromatic images, we found that
the ring radius is typically between 4 and 18 m. We constructed ring filters with radii in
this range, see Figure 4.8.

A mask representing the agricultural fields was applied to the image before further
processing took place. We then performed contrast enhancement of the images. Template
matching was performed by letting the binary filter “slide” across the contrast-enhanced
image. For each image position of the template the filter response was recorded. The
result was an image of where the locations where the templates match well will have

18
relatively high or low
values. High values
indicate a match with a
bright ring, whereas
low values indicate a
match with a dark ring.

The next step was

feature extraction. For
each candidate ring
position we extracted a
surrounding sub-image Figure 4.8. Binary (left) and intensity (right) ring templates
from the original image.
Different features calculated from this sub-image were used to determine whether to
reject or accept the candidate as a ring. These criteria were derived from investigation of
feature values for known rings. Perhaps the most important feature is ring correlation,
which is a measure of how well the sub-image resembles the intensity ring template in
Figure 4.8.

Figure 4.9 and 4.10 display some of the sub-images resulting from ring detection. Table
4.1 displays the detection rates.

Figure 4.9. Sub-images of falsely detected rings

Figure 4.10. Sub-images of detected rings

19
 Number of Number of Number of
Image known correct false
rings/disks detections detections
1 1 1 12
2 1 0 6
3 1 1 6
4 1 1 6
5 1 0 8
6 1 1 7
7 1 1 3
8 2 1 10
9 4 4 1
10 2 2 2
11 4 4 6
TOTAL 19 16 67

20
5 CultSearcher from an archaeologist’s viewpoint
Vestfold County and the Museum of Cultural History (KHM) in Oslo have been involved
in the project since 2003. Vestfold County was in the period 2003–2005 represented by
archaeologist Trude Aga Brun, and since 2005 also included archaeologist Christer
Tonning. Archaeologist Lars Gustavsen has represented Museum of Cultural History
throughout the project period. The archaeologists have contributed in testing and
commenting, and thereby shaping the CultSearcher software prototype as a potential tool
for future cultural heritage management in agricultural landscapes throughout the country.

The aim of the exercise from the archaeologist's point of view was twofold: Firstly, it
aimed to test whether the software would be capable of detecting actual archaeological
features in satellite imagery. Secondly, it was aimed at testing the suitability and
functionality of the software for use in a cultural heritage management environment.

In order to test the functionality of the software, two geographically separate areas with
suitable satellite imagery had to be selected. These had to consist mainly of cultivated
landmass and have a reasonably dense population of already recorded scheduled
monuments. Further to this, the imagery covering the areas had to have little or no cloud-
cover and the images had to have been taken at the right time of year.

The areas selected consisted of images recorded around an area near Tønsberg in
Vestfold County and an area in the vicinity of Gardermoen airport in Akershus County.
Archival satellite imagery from both areas was inspected in order to select imagery
without cloud cover and within a time frame which would be favourable for crop- or soil-
mark detection. Both areas were covered by panchromatic and multispectral imagery.
The datasets differed, however, in that the Tønsberg coverage consisted of imagery from
the IKONOS satellite whereas the Gardermoen images were recorded by the Quickbird
satellite.

For the sake of convenience it was decided that the Vestfold imagery was to be inspected
by an archaeologist from Vestfold County, whereas the Akershus imagery would be dealt
with by an archaeologist from the Museum of Cultural History in Oslo.

5.1 Testing CultSearcher on Quickbird images from the Gardermoen area

5.1.1 The test area

The Akershus area partly covers the municipalities of Ullensaker and Nannestad, an area
amounting to a total of some 600 km2. Archaeological investigations have revealed that
the area has been settled since the Neolithic period (about 4000–1800 BC), and the
Askeladden database of scheduled monuments lists over 900 sites of varying types and
dates. Of these, approximately 180 consist of burial mounds or cairns dating from the
Bronze Age to the late Iron Age. One of the more spectacular monuments in the area is
Raknehaugen, a colossal tumulus measuring some 90 m in diameter and 15 m in height.
Although its function has never been established, C14 dates suggest that it was

21
constructed about 500 AD. In addition, the area is known for an extensive network of
prehistoric roads as well as a number of medieval churches.

In addition to this dense concentration of archaeological sites, the area is characterised by

easily cultivated moraine landscapes, and is considered a typical eastern Norwegian
cultural landscape. The archival imagery selected for this part of the test was recorded on
27 July 2003, a time of year which should be favourable for crop-mark detection. It is the
time of year when crops are in the process of ripening and thus turning yellow. The
combination of these factors makes the area particularly suitable for detecting potential
crop marks.

5.1.2 The test

The first part of the test included selecting and ordering appropriate imagery from the
available preview images. A convenient aspect of the Quickbird imagery is that it is
possible to select areas by using irregular polygons. These can be based on already
available datasets, or can be generated on-the-fly. Thus, it is possible to exclude forested
or urban areas from the ordered data. As the imagery of the study area were not geo-
referenced to the same accuracy as the available datasets, it was necessary to select areas
manually. This was done by importing the images to ESRI ArcGIS, where geo-specific
polygons could be drawn over the areas of interest.

When the imagery had become available, it was possible to proceed with the second
phase of the software testing. Testing the CultSearcher prototype software involved a
series of different steps. Firstly, the selected images had to undergo visual inspection by
the archaeologists in order to pick out potential archaeological targets. This involved
carefully going through each individual patch of cultivated land in the images trying to
identify crop and/or soil marks. As a guide to where crop marks might be visible, data
from the Askeladden database in the form of points were exported from the database and
overlaid the satellite imagery.

When suitable areas had been selected based on these criteria, the images had to be
divided into sub-images of the original image. This was done in order for the software to
be able to handle the substantial amounts of data present in the satellite images. These
sub-images then had to be imported into the CultSearcher software, and masks had to be
created over the areas on which the software was to be tested. When this was done the
programme was executed, and contrasting features were detected as previously described.

5.1.3 Preliminary observations

Following the preliminary testing it is my opinion that the software as it stands is
unsuitable for use by personnel without specific knowledge of satellite imagery and/or
GIS-related software. Furthermore, as the ring algorithm had not yet been implemented in
this phase of the testing, the software would only pick out areas with amorphous features.
These are features that cannot be positively identified as archaeologically significant
without investigation in the field.

22
Another obstacle for implementing this software in a cultural heritage environment is the
user interface and the way the software stores data. The software is currently far too
cumbersome and will have to become much more automated in order to work in a normal
administrative setting. The most time-consuming aspect of using the software is that of
having to select areas of interest and create masks manually. This, in fact takes longer
than the actual computational analysis of the image, and will have to become a more
automated process.

An ideal process would be as follows: A satellite image is loaded into the software. The
software then automatically identifies areas of interest. This would have to identify
cultivated landmass and exclude forested and inhabited areas. The software should then
split the original image into smaller images which can be analysed more efficiently by
'normal' computers. Following the analysis of the area, the software should then give the
user the option to export the results to familiar vector-based formats such as shape or dxf
for further use in GIS/CAD packages common in cultural heritage management.

It is clear that this software has the potential for further development. From the visual
inspections of the satellite imagery it became evident that archaeological features can in
fact be seen from space, in the form of ring-shaped crop marks. Therefore, with a suitable
algorithm for detecting these, CultSearcher will be of great value for archaeologists in the
field of cultural heritage management.

5.2 Testing CultSearcher on Ikonos images from Vestfold County

5.2.1 The test area

In the first edition of CultSearcher a considerable amount of work had been put into the
software to enable the system to detect amorphous objects in Quickbird satellite images.
In Vestfold County a Quickbird image of the southern parts of the river Lågen, was used
to train CultSearcher to detect amorphous objects. In this Quickbird image an Iron Age
grave field located at Odberg farm was the central target, and the surrounding fields were
subject to intense investigation and search for other similar or related archaeological
objects not visible in situ.

5.2.2 Testing various versions of CultSearcher

In this phase of the project, with CultSearcher detecting amorphous objects where
archaeologists did not have any information about the sites detected, it became clear that
we had to re-evaluate central issues in developing CultSearcher. Detections of amorphous
objects where no verification – either by archaeological survey, excavation or archive
data – is clearly of little value and could not bring the functionality of the software further.

Instead of pursuing the search for amorphous objects in unknown territory, the focus was
shifted to extending CultSearcher to be able to detect ring-shaped objects where Vestfold
County and KHM had information from archives, surveys or excavations on where
archaeological objects had been situated, and where possible traces of them could be
detected. In this way we may start from a known point of origin and it would be possible

23
to train CultSearcher to detect similar features in unknown landscapes. A ring ditch
surrounding an Iron Age grave mound is a fairly common archaeological feature
especially in Vestfold County, but also in Norway in general. The subject matter is
plentiful and therefore especially well suited for training CultSearcher in this early phase.

5.2.3 Remarks
The user interface in CultSearcher and the process from creating masks and importing
satellite images to extracting the final interpretation of detections is still somewhat
complicated. Many similar processes, producing files with similar names can be
confusing for the beginner using the software. A clearer structure where archaeological
definitions are properly incorporated in the graphical user interface and the output data is
necessary for widespread use of the software amongst archaeologists in cultural heritage
management.

CultSearcher is clearly still a prototype software, but with significant prospects. It can
offer archaeologists a better view of what to expect when prospecting and excavating
agrarian landscapes. CultSearcher may offer greater success in actually locating
archaeological remains, and better possibilities of understanding the bigger picture of the
prehistory of landscapes.

In the future Vestfold County would like to broaden the variety of archaeological remains
to be detected in CultSearcher, e.g. houses, cooking pits, walls, wall ditches and roads.
The learning process of CultSearcher is still in progress, but we have high expectations
and great belief in the ongoing systematic approach where CultSearcher should be a
powerful tool in the ever growing archaeological toolbox.

24
6 Plans for the future
A software prototype for the detection of potential cultural heritage sites in high-
resolution satellite images has been developed. The Directorate for Cultural Heritage’s
aim is that the system will become a key operational tool for cultural heritage
management nationwide. Furthermore, it is hoped that this approach will reduce the
number of excavations in order to establish in situ protection and management of cultural
heritage sites. Additionally, a central perspective of using satellite technology for cultural
heritage management is the potential of reporting on national goals for the environmental
policy.

The prototype system CultSearcher was presented at an international seminar held by the
Directorate for Cultural Heritage and the Norwegian Space Center in Oslo 9–10 January
2008. The reactions of the audience, with representatives including the Norwegian county
administrations and researchers from several countries, were positive. Nonetheless, it
became clear that the system still has some challenges to overcome.

At present it is obvious that there is still a long way to go to satisfy the demands for good
management, as stated in Chapter 5 of this report. It is therefore crucial to continue with
the development of suitable algorithms for detecting sites of interest. Furthermore, it is
important that more users, such as archaeologists in the county administrations,
participate in this development work. CultSearcher will only be accepted if the end users
see the value of the tool.

In the near future (2008) we will run the system for all the agricultural areas in Vestfold
County. This will give us a broader overview of the obstacles and important experience
for further development of the algorithms. For visualisation purposes, we want to create
GIS maps for the agricultural areas in Vestfold County showing graded potential
locations of cultural heritage sites. We believe that those maps will underline the
management aspect within land-use planning purposes. If CultSearcher can help create
maps showing potential locations, we can use this information at an early stage in the
planning processes, e.g. environmental impact assessments, thus eventually contributing
to a more far-sighted planning practice.

Since we understand that there is international interest in our methodological approach,

we want to be active in the national and international research arenas. The Directorate
encourages all project partners to engage in upcoming research applications, including
the cultural heritage management contributing with the user perspective. Participation in
national and international research programmes also enables the cultural heritage
management to focus on problems which fall outside the current work on developing an
operational tool.

If CultSearcher is able to satisfy the demands of the cultural heritage management on a

regional level, the Directorate will work towards covering all the agricultural areas in
Norway with satellite images, resulting in an overview of potential locations of cultural

25
heritage sites nationwide. To challenge ourselves – we suggest the year 2013 as a
possible milestone.

26
7 References
Amlien, J., Holden, M., Larsen, S.Ø. and Solberg, R (2007). CultSearcher - Software
Guide. Research Note, no. SAMBA/39/07, Norwegian Computing Center.

Aurdal, L., Eikvil, L., Koren, H. and Loska, A. (2006). Semi-automatic search for
cultural heritage sites in satellite images. In proceedings of From Space to Place, 2nd
International Conference on Remote Sensing in Archaeology, 4-7 December 2006, Rome.

Aurdal, L., Solberg, R. and Lous, J. (2005). Computer assisted detection and
classification of possible cultural heritage sites: User requirements and conceptual
design. Research Note, no. SAMBA/11/05, Norwegian Computing Center.

Grøn, O; MacPhail, R.; Christensen, F.; Loska, A.; Aurdal, L.;·Solberg,R.; Lous, J.
(2005). Locating invisible cultural heritage sites in agricultural fields. Development of
methods for satellite monitoring of cultural heritage sites – report 2004. University
College London, Norsk Regnesentral, Norsk Romsenter, Riksantikvaren.

Grøn, O., Aurdal, L., Christensen, F., Tømmervik, H. and Loska, A. (2004). Locating
invisible cultural heritage sites in agricultural fields: Evaluation of methods for satellite
monitoring of cultural heritage sites – Results 2003. Research Report, March 2004,
NIKU, Norsk Romsenter, Riksantikvaren.

Grøn, O. & Loska, A. (2002). Development of methods for satellite monitoring of cultural
heritage sites. Research Report, December 2002, NIKU, Norsk Romsenter,
Riksantikvaren.

Trier, Ø.D., Larsen, S.Ø. and Solberg, R. (2008). Detection of circular patterns in high-
resolution satellite images of agricultural land with CultSearcher. Research Note, no.
SAMBA/16/08, Norwegian Computing Center.

Niblack, W. (1986). An Introduction to Digital Image Processing. Prentice Hall.

27
Contact information
Riksantikvaren, RA (Norwegian Directorate for Cultural Heritage):
Anke Loska, Riksantikvaren, P.O. Box 8196, N-0034 OSLO, Norway; http://www.ra.no;
[email protected]; Phone +47 2294 0400.

Norsk Regnesentral, NR (Norwegian Computing Center):

Rune Solberg, Norwegian Computing Center, P.O. Box 114 Blindern, N-0314 Oslo,
Norway; http://earthobs.nr.no; [email protected]; Phone: +47 2285 2500.

Kulturhistorisk museum, KHM (Museum of Cultural History, University of Oslo):

Lars Gustavsen, Dokumentasjonsseksjonen, P.O. Box 6762 St. Olavsplass, N-0130 Oslo,
Norway; http://www.khm.uio.no; [email protected]; Phone: +47 2285 9716.

Vestfold fylkeskommune, VFK (Vestfold County Administration):

Christer Tonning, Cultural Heritage, Vestfold County Administration, Svend Foynsgate 9,
N-3128 Tønsberg, Norway; [email protected]; Phone: +47 9597 2260.

28