Revised: 2 August 2018 Accepted: 7 October 2018

DOI: 10.1002/arp.1730

RESEARCH ARTICLE

Object‐based image analysis: a review of developments and

future directions of automated feature detection in landscape
archaeology
Dylan S. Davis

Department of Anthropology, The

Pennsylvania State University, University Park, Abstract
PA, USA Object‐based image analysis (OBIA) is a method of assessing remote sensing data that
Correspondence
uses morphometric and spectral parameters simultaneously to identify features in
Dylan S. Davis, Department of Anthropology,
The Pennsylvania State University, University remote sensing imagery. Over the past 10–15 years, OBIA methods have been intro-
Park, PA 16802, USA.
duced to detect archaeological features. Improvements in accuracy have been
Email: [email protected]
attained by using a greater number of morphometric variables and multiple scales of
analysis. This article highlights the developments that have occurred in the application
of OBIA within archaeology and argues that OBIA is both a useful and necessary tool
for archaeological research. Additionally, I discuss future research paths using this
method. Some of the suggestions put forth here include: pushing for multifaceted
research designs utilizing OBIA and manual interpretation, using OBIA methods for
directly studying landscape settlement patterns, and increasing data sharing of
methods between researchers.

KEY W ORDS

automated feature extraction, landscape analysis, machine learning, object‐based image analysis,
pattern recognition, remote sensing

1 | I N T RO D U CT I O N focus on European localities, but fewer focus on the Americas,

Asia, Africa, or island regions. Furthermore, most archaeological
Researchers in many fields – including computer science and geogra- publications using OBIA methods are identifying potential sites,
phy – have adopted machine learning algorithms to process remote but they are not addressing potential settlement patterns that
sensing imagery (see Mountrakis, Im, & Ogole, 2011). In the late emerge from their results. This is important because future
1990s and early 2000s, a form of machine learning known as research should use these methods to answer archaeological
object‐based image analysis (OBIA) was developed (Blaschke, questions concerning populations, socio‐political organization, and
2010), but only recently have archaeologists utilized these methods past peoples at large.
(e.g. De Laet, Paulissen, & Waelkens, 2007; Menze, Ur, & Sherratt, This article serves as a review of object‐based machine learning
2006). In the last ~15 years, archaeologists have used a variety of methods that archaeologists have applied in landscape‐scale remote
OBIA techniques that are highly successful in extracting features of sensing analysis – including aerial and spaceborne data. I will detail
interest from large‐scale datasets at faster rates and lower costs the progress that has been made with these techniques as well as
than manual processing (Bennett, Cowley, & De Laet, 2014, 897). the avenues archaeologists are yet to travel. I begin by reviewing
Yet, there is much more that these methods can do to advance the basic concepts of OBIA and how it operates. I follow with a
our understanding of the human past. comprehensive summary of archaeological work that has been con-
Today, there have been a number of significant studies using ducted using OBIA, paying particular attention to the successes
OBIA methods within archaeological contexts, but these studies and shortcomings of these studies. Then, I discuss possible future
are not evenly distributed geographically (see Table 1). Many directions of OBIA and computational archaeology. I illustrate that

Archaeological Prospection. 2018;1–9. wileyonlinelibrary.com/journal/arp © 2018 John Wiley & Sons, Ltd. 1
2 DAVIS

TABLE 1 Archaeological studies using object‐based image analysis sensing analysis in archaeology). Pixel‐based approaches deal with
(OBIA) by geographic region the classification of individual pixels in an image into different
Region OBIA studies categories corresponding to unique landscape features. Researchers

Europe Bescoby, 2006; Cerrillo‐Cuenca, 2017; D'Orazio,

have used these methods within many fields for a variety of purposes
Palumbo, & Guaragnella, 2012; Guyot et al., 2018; including: land cover and vegetation classification, mapping urban
Magnini et al., 2017; Schneider et al., 2015;
expansion, and measuring surface temperatures (see Jensen, 2007).
Sevara, Pregesbauer, Doneus, Verhoeven,
& Trinks, 2016; Traviglia & Torsello, 2017; As such, pixel‐based classification is quite useful. However, when
Trier & Pilø, 2012; Trier et al., 2009; compared to object‐based approaches, OBIA methods are more
Trier et al., 2015; Verhagen & Drăguţ, 2012;
Zingman, Saupe, Penatti, & Lambers, 2016 accurate for detecting archaeological features (see De Laet et al.,
North America Davis et al., In Press; Davis et al., 2018; 2007; De Laet, Music, Paulissen, & Waelkens, 2008; Sevara &
Johnson & Ouimet, 2014; Kvamme, Pregesbauer, 2014).
2013; Riley, 2009; Witharana et al., 2018
OBIA methods, in contrast to pixel‐based methods, identify fea-
South America Lasaponara & Masini, 2018
tures using multiple variables. These include pixel value, object shape,
Asia De Laet et al., 2007, 2008; De Laet, Paulissen,
Meuleman, & Waelkens, 2009; Harrower, textural information, neighbourhood analysis, and geographic context
Schuetter, McCorriston, Goel, & Senn, (Blaschke, 2010, 3; Blaschke et al., 2014). By utilizing multiple param-
2013; Jahjah et al., 2007; Lasaponara
& Masini, 2018; Menze et al., 2006;
eters simultaneously, OBIA is well suited for identifying features that
Menze, Mühl, & Sherratt, 2007; are small, structurally homogeneous, and display differences with local
Menze & Ur, 2012; Schuetter, et al.,
topography (Davis, Sanger, & Lipo, 2018). OBIA builds on longstanding
2013; Van Ess et al., 2006;
Wang et al., 2017 practices of remote sensing analysis including segmentation, edge
Pacific Islands Freeland et al., 2016 detection, and classification (Blaschke, 2010, 3; see Kumar, Raj Kumar,
& Reddy, 2014; Weng, 2010, for reviews of different types of seg-
mentation and classification). As such, some have considered OBIA
while archaeologists have just begun to apply OBIA to research to be one of the greatest achievements in image processing of the
questions, the method offers unparalleled advantages that should twenty‐first century (Arvor, Durieux, Andrés, & Laporte, 2013). One
be fully taken advantage of by future archaeological research. limitation of OBIA is that it requires very high‐resolution datasets to
work effectively (Blaschke et al., 2014, 181). However, as the spatial
resolutions of remotely sensed data have improved, the accuracy
2 |
OBJECT‐BASED IMAGE ANALYSIS and use of OBIA techniques have also increased (Hay et al., 2005).
(O BIA)

OBIA began to rise in popularity in the early twenty‐first century, and 3 | OBIA AND MACHINE LEARNING IN
since that time, uses of these methods have sharply increased ARCHAEOLOGY
(Blaschke, 2010). Blaschke (2010) attributes this rise to the develop-
ment of a software called eCognition (Trimble, 2016). In addition to Object‐based analysis of remote sensing data has only been exten-
eCognition, several open‐source platforms have been developed for sively utilized by archaeologists for about 15 years. The number of
OBIA analysis [e.g. GEODMA (Körting, Garcia Fonseca, & Câmara, peer reviewed publications using these methods within archaeological
2013), InterIMAGE (InterIMAGE, 2009), Grass GIS (GRASS Develop- contexts is small (< 40) but growing (see Table 2). Additionally, a great
ment Team, 2018), also see Knoth & Nüst, 2017]. deal of work has been presented on the use of OBIA at archaeological
In its most basic definition, object‐based analysis encompasses conferences.
‘image‐processing techniques that when applied either result in the Beginning in the first decade of the twenty‐first century,
segmentation (i.e. partitioning) of an image into discrete non‐ researchers began to implement object‐based computer algorithms
overlapping units based on specific criteria, or are applied to define to detect archaeological features in a systematic fashion. The first
specific multiscale characteristics—from which segmentation may then archaeological research implementing OBIA was primarily concerned
be based’ (Hay, Castilla, Wulder, & Ruiz, 2005, 340). Recently, remote with identifying large‐scale linear features. For example, Bescoby
sensing literature has used the term GEOBIA to refer to those applica- (2006) used a mathematical function known as a Radon transform
tions of OBIA to Earth remote sensing imagery (Blaschke et al., 2014; (which can determine the most common alignment and orientation
Hay & Castilla, 2008). GEOBIA therefore constitutes a majority of of features within an image) and segmentation procedures to detect
OBIA applications within archaeology. linear Roman structures in satellite imagery. Within a few years, more
In a discussion of automated extraction methods used within publications began to emerge using OBIA methods (e.g. De Laet et al.,
archaeology, one cannot ignore the multitude of studies that have 2007; Jahjah, Ulivieri, Invernizzi, & Parapetti, 2007; Van Ess et al.,
used pixel‐based classification (e.g. Bennett, Welham, Hill, & Ford, 2006). All of these studies focus primarily on the detection of archae-
2012; Campbell, 1981; Custer, Eveleigh, Klemas, & Wells, 1986; ological deposits, but Jahjah et al. (2007) also look at how OBIA tech-
Drager, 1983; Kirk, Thompson, & Lippitt, 2016; Lasaponara, Leucci, niques can monitor sites, document their preservation levels (also see
Masini, & Persico, 2014; Lasaponara & Masini, 2007; Meredith‐ Van Ess et al., 2006), and enhance the digitization of archaeological
Williams et al., 2014; also see Lambers, 2018, for a review of remote data acquired from remote sensing sources. As the resolution of
DAVIS 3

TABLE 2 Outline of developments of object‐based image analysis (OBIA) as a method in archaeology from 2000 to the present

Time period Advances Limitations

2000–2010 • Segmentation and mathematical algorithms (Bescoby, 2006) • Most studies use two‐dimensional (2D) aerial
• Edge detection for roadway identification (De Laet et al., 2007) and satellite imagery, not three‐dimensional
(3D) topographic datasets such as LiDAR
• Morphometric variables used for classification include: (Menze et al., 2006 is the exception).
• Shape (De Laet et al., 2007; Menze et al., 2006) This limits the variables that can be
• Compactness and smoothness (De Laet et al., 2007) used to identify features.
• Colour/pixel value (Jahjah et al., 2007) • Low number of morphometric criteria
• Neighbourhood analysis (De Laet et al., 2007, 2008, 2009) were inefficient at capturing morphological
diversity of certain feature types
• Elevation (Menze et al., 2006)
• High number of false‐positive and
• Pattern recognition (template matching) is implemented as false‐negative results (e.g. De Laet et al.,
an automatic detection method (Trier et al., 2009) 2007, 2008, 2009; Menze et al., 2006,
2007; Trier et al., 2009)
2010–2015 • Use of LiDAR and 3D datasets becomes prevalent • Spatial resolution of remote sensing data (> 1 m)
• Morphometric variables used for classification include: sometimes prevent accurate detection of small
• Elevation, slope, and curvature (Schneider et al., 2015; deposits (Verhagen & Drăguţ, 2012)
Verhagen & Drăguţ, 2012) • Many of these methods have significant issues
• Nearest neighbour analysis, size, shape, and circularity with false‐positive
(Harrower et al., 2013; Scheutter et al., 2013) and false‐negative results (e.g. Kvamme, 2013;
Harrower et al., 2013; Scheutter et al., 2013;
• Hillshade and topographic position index Schneider et al., 2015; Trier & Pilø,
(Schneider et al., 2015) 2012; Trier et al., 2015)
• Principle component analysis (Chen, Comer, • Some methods cannot detect features
Priebe, Sussman, & Tilton, 2013) (or present false positives) that are located in
• Orientation and topographic contours close proximity to certain types of topographic
(D’Orazio et al., 2012; Figorito & Tarantino, 2014) anomalies (e.g. D'Orazio et al., 2012; Trier et al.,
• Eccentricity (Figorito & Tarantino, 2014) 2015; Trier & Pilø, 2012)
• Volume (Menze & Ur, 2012)
• Pattern recognition (template matching) continues
to be utilized (Kvamme, 2013; Schneider et al., 2015;
Trier et al., 2015; Trier & Pilø, 2012)
• Increased accuracy achieved by:
• Use of statistical classifiers (Chen et al., 2013;
Trier et al., 2015)
• Using multiple datasets to cross‐reference automated
results from different sources (Trier et al., 2015)
• Using multitemporal datasets to account for possible
seasonal lapses in visibility in remotely sensed
imagery (Menze & Ur, 2012)
2015–2018 • Co‐opting of hydrological depression analyses for mound detection • False‐positive and false‐negative results
(Davis et al., In Press; Freeland et al., 2016) (Freeland et al., 2016; Schneider et al.,
• Morphological variables used for classification include: 2015; Trier et al., 2015; Witharana et al., 2018)
• Circularity (Davis et al., 2018; Freeland et al., 2016; • Lack of temporal control during feature detection
Witharana et al., 2018) (Traviglia & Torsello, 2017)
• Rectangularity (Zingman et al., 2016) • Small sample sizes for pattern recognition
• Area (Davis et al., 2018; Magnini et al., 2017; (Davis et al., 2018; Wang et al., 2017;
Witharana et al., 2018) Zingman et al., 2016)
• Length and width (Magnini et al., 2017; Toumazet,
Vautier, Roussel, & Dousteyssier, 2017)
• Size (Cerrillo‐Cuenca, 2017; Davis et al., 2018;
Zingman et al., 2016)
• Curvature (Cerrillo‐Cuenca, 2017)
• Edge detection (Traviglia & Torsello, 2017;
Witharana et al., 2018; Zingman et al., 2016)
• Elevation (Davis et al., 2018; Guyot et al., 2018)
• Multiscalar analysis with multiple datasets are incorporated
to cross‐validate results at small‐to‐large scales
(Guyot et al., 2018; Witharana et al., 2018)
• Pattern analysis continues to be used
(Davis et al., 2018; Trier et al., 2015; Wang et al., 2017)
4 DAVIS

remote sensing data improved, smaller features were soon the subject greatly enhance our ability to detect above‐ground archaeological
of detection via automated processes (e.g. Magnini, Bettineschi, & De structures.
Guio, 2017; Wang, Hu, Wang, Ai, & Zhong, 2017; also see Beck, Philip, Trier and Pilø (2012) show how pattern recognition via template
Abdulkarim, & Donoghue, 2007). matching can incorporate many of these morphometric properties into
The research conducted prior to 2010 illustrates the first attempts the classification of topographic data (e.g. LiDAR). The procedure
at defining archaeological deposits as objects in a manner that com- involves the creation of samples of known features of interest (i.e.
puters can understand and replicate via segmentation and classifica- templates) from digital elevation models (DEMs) using different scales
tion procedures. Variables including shape, compactness, texture, and resolutions. The template matching algorithm is then conducted
and colour are all implemented as parameters for detecting likely on each DEM and the computer extracts identifications that overlap
archaeological features (De Laet et al., 2007; Jahjah et al., 2007; see between the different scales – thereby acting as a cross‐check
Table 2). However, most of these studies use only two‐dimensional between each iteration of the algorithm. The method assesses each
(2D) satellite imagery, and three‐dimensional (3D) topographic data feature for its degree of statistical similarity to the templates and
[such as LiDAR (light detection and ranging)] is not analysed using assigns a corresponding confidence interval. The final results are then
OBIA [Menze et al. (2006) is one exception]. Furthermore, these early field‐tested by archaeologists. This procedure – which several
studies suffer from high rates of false‐positive identifications, which is researchers have used in different variations (e.g. Kvamme, 2013;
a result of the quality of the data used and the variables incorporated. Schneider, Takla, Nicolay, Raab, & Raab, 2015; Trier, Larsen, &
Implementing OBIA with a greater number of variables (including mul- Solberg, 2009; Trier, Zortea, & Tonning, 2015) proves successful in
tiple scales of analysis) and using higher‐resolution datasets improves the identification of previously detected and undetected archaeologi-
accuracy for archaeological prospection (e.g. Guyot, Hubert‐Moy, & cal structures. As with earlier studies, however, template matching is
Lorho, 2018; Sevara & Pregesbauer, 2014, 142; Witharana, Ouimet, limited by its number of false‐positive and false‐negative results.
& Johnson, 2018). Subsequent research has implemented a slew of new variables
By the beginning of the 2010s there is an increase in archaeolog- including topographic measurements such as hillshade, slope, and
ical studies analysing LiDAR and topographic datasets – in addition to topographic openness (see Table 2). The results of these studies indi-
2D satellite and aerial imagery – using OBIA procedures (e.g. Trier & cate a positive correlation between the number of factors accounted
Pilø, 2012; Verhagen & Drăguţ, 2012). By incorporating 3D data, the for during OBIA procedures and accuracy. However, if the parameters
detection of archaeological features becomes easier, as researchers chosen do not match the features that are being sought after then the
can now incorporate topographic information in multiple dimensions. algorithm will not work. As such, an expert knowledge of the study
For example, Verhagen and Drăguţ (2012) use elevation, slope, and area is an essential prerequisite for using automated detection
curvature as parameters for segmentation and classification of land- methods.
forms. Although their method is not perfectly accurate – a large part The increase in studies post‐2010 sees a slight diversification in
of which is due to the resolution of the datasets used (5 m) and the the use of OBIA methods, but as Table 3 shows, most studies use it
number of variables included in the segmentation procedure – for the sole purpose of automating the detection of archaeological
Verhagen and Drăguţ (2012) illustrate how incorporating 3D morpho- features (e.g. Davis et al., 2018; Sevara & Pregesbauer, 2014; Trier
logical and morphometric variables, in addition to 2D profiles, can et al., 2015; Trier & Pilø, 2012; Verhagen & Drăguţ, 2012). Some

TABLE 3 Publications using object‐based image analysis (OBIA) as a primary method organized by their research goals
Number of
Research goal publications References

Identification 28 Bescoby, 2006; Cerrillo‐Cuenca, 2017; Chen et al., 2013;

D’Orazio et al., 2012; Davis et al., 2018, in press;
De Laet et al., 2007, 2008, 2009; Figorito & Tarantino,
2014; Freeland et al., 2016; Guyot et al., 2018;
Harrower et al., 2013; Jahjah et al., 2007; Kvamme,
2013; Menze et al., 2006, 2007; Schneider et al., 2015;
Schuetter et al., 2013; Sevara & Pregesbauer, 2014;
Sevara & Pregesbauer, 2014; Toumazet et al., 2017;
Traviglia & Torsello, 2017; Trier et al., 2009, 2015;
Verhagen & Drăguţ, 2012; Witharana et al., 2018;
Zingman et al., 2016
Preservation/monitoring 6 Lasaponara & Masini, 2018; Magnini et al., 2017;
Sevara & Pregesbauer, 2014; Trier & Pilø, 2012;
Van Ess et al., 2006; Wang et al., 2017
Mapping/digitization of archaeological features 3 Lasaponara & Masini, 2018; Sevara & Pregesbauer,
2014; Witharana et al., 2018;
Analysis of populations, social organization, settlement patterns, etc. 4 Cerrillo‐Cuenca, 2017; Cordero Ruiz et al., 2017;
Freeland et al., 2016; Menze & Ur, 2012

Notes: Some publications fall into multiple categories, and as such are listed multiple times. There is a total of 35 sources that have been included in this
table.
DAVIS 5

researchers, however, are using OBIA to protect and monitor sites at features. Part of their discussion revolves around a distrust of these
risk of destruction (e.g. Lasaponara & Masini, 2018; Magnini et al., methods, and they state:
2017; Schneider et al., 2015; Trier & Pilø, 2012; Wang et al., 2017)
The reluctance to adopt automated feature extraction …
and to develop more complete maps of archaeological activity to con-
is motivated by a combination of technological and social
duct further analysis of settlement patterning and sociopolitical orga-
factors. On the technological side, machine learning
nization (e.g. Cerrillo‐Cuenca, 2017; Cordero Ruiz, Cerrillo Cuenca, &
approaches to automation remain in their infancy.
Pereira, 2017; Freeland, Heung, Burley, Clark, & Knudby, 2016).
Automatic feature extraction for archaeological
The most recent archaeological uses of OBIA and machine learn-
materials is still developing and has yet to match the
ing yield highly accurate results (Freeland et al., 2016; Guyot et al.,
efficiency of automatic feature extraction for targets
2018; Lasaponara & Masini, 2018; Wang et al., 2017). Freeland et al.
with consistent appearance or for features in uniform
(2016) demonstrate the first use of hydrological depression algorithms
environments. (Opitz & Herrmann, 2018, 30)
for archaeological mound detection. In this instance, an inversed DEM
was created and processed through an algorithm that looks for topo- The claim that these methods are still new and evolving is very much
graphic depressions, effectively identifying and mapping mound fea- true, as this article indicates. Regardless, the infancy of the method is
tures (also see Davis, Lipo, & Sanger, in press). The work of Guyot not a reason to stop developing and improving its ability to discern
et al. (2018) is a strong case for the use of automated object extraction information of archaeological significance. OBIA and similar methods
methods within archaeology, as their multiscalar algorithm success- are imperfect and cannot replace manual evaluation completely, but
fully identified over 2000 Neolithic burial mounds, while false posi- at the same time, biases in knowledge by data analysts limit the accu-
tives (n = 41) and false negatives (n = 46) were minimal. The lesson racy of manual procedures and can lead to omission error (Bennett
from this latest research are simple: landscape level archaeological et al., 2014; Gheyle et al., 2018). It can never be our goal to completely
prospection algorithms must be multiscalar. Furthermore, topographic automate the archaeological process, and to attempt such a feat
data are invaluable for the automated identification of archaeological would be a fool's errand. Nevertheless, improving automated methods
deposits. With very‐high resolution datasets and the accuracy to assist in the detection of archaeological deposits is not only an
improvements acquired in recent research, OBIA shows promise for exciting avenue for future research, but also a necessary task.
1
highly accurate automated prospection. Coastal and island regions that are under threat of destruction by
climate change and rising sea levels cannot ever be fully surveyed
using traditional means before their records are severely damaged. It
3.1 | Limitations and criticisms
is therefore imperative to document as much of these areas as we
can before they are lost. By using OBIA and similar methods, we can
Despite the many successes of OBIA methods within archaeology,
conduct systematic surveys of entire areas and document landscapes
there are many who are skeptical of the feasibility of automated
efficiently. Thus, it is essential to utilize these techniques to study
detection algorithms – specifically for large‐scale landscape analysis
the archaeological record in a relatively complete form rather than lim-
(e.g. Casana, 2014; Hanson, 2010; Parcak, 2009). Parcak (2009,
iting ourselves to small sample sizes of information.
110) claims that automated archaeological site detection is impossi-
Despite the benefits offered by OBIA, it is still far more common
ble because every archaeological project is dependent on local vari-
for archaeologists to use manual interpretation methods rather than
ables. But local variables are precisely what OBIA can take into
semi‐automatic means (Quintus, Day, & Smith, 2017, 352; also see
consideration when analysing remote sensing data, and regionally
Casana, 2014). Many researchers echo the earlier sentiments of
specific algorithms are essential for the success of automated
Parcak (2009) by claiming that automated methods cannot account
prospection (see Davis et al., 2018). Parcak (2009,110) goes on to
for the wide range of variability in the archaeological record. However,
state that computers cannot pick up on the same subtleties in
who is to say that one should only use one single automated method
remotely sensed data as humans can by eye. However, the very fact
to scan an entire study area? Why not use a multitude of different
that recent studies using automated means have detected sites that
algorithms to search for different parts of the record and then go
manual analysis has overlooked directly challenges this claim (e.g.
through all the results by hand to fill in things that OBIA missed (sensu
Davis et al., 2018; Witharana et al., 2018).
Bennett et al., 2014)? By using automated detection first, we can be
In discussing the latest state of remote sensing research within
sure that the entire study area is surveyed systematically without
archaeology, Opitz and Herrmann (2018) devote some of their atten-
any lapses. Then by conducting a manual analysis, expert knowledge
tion to the methods involving automated detection of archaeological
can assess the results and potentially identify nearby features that
1
Although this article has focused exclusively on the use of OBIA for large‐scale the automated method overlooked.
remote sensing data such as satellite imagery and LiDAR (GEOBIA), this method Casana (2014) uses ‘brute force’, or manual extraction methods to
has also been used for other types of image analysis in archaeology. OBIA has
survey an area covering 300 000 km2. This process took approxi-
been successful in classifying artefacts and features into statistically significant
types (e.g. Lamotte & Masson, 2016; Ozawa, 1978), studying site formation pro- mately 3–4 years. Using automated methods [which Casana (2014)
cesses (e.g. Sanger, 2015), testing the mineralogical classification of artefacts (e. attempted and stated to be successful], this process could have been
g. Aprile et al., 2014; Hein et al., 2018; Hofmann et al., 2013), and researchers
sped up considerably. Quitnus et al. (2017) also illustrate the impor-
have used it to investigate ground‐based remote sensing data (e.g. Pregesbauer,
Trinks, & Neubauer, 2014). As such, there is much to gain from OBIA methods tance of manual evaluation, but highlight the fact that manual process-
within archaeology that goes beyond landscape‐level analysis. ing is imperfect, as there are still many false‐positive and false‐
6 DAVIS

negative identifications present using this approach. In the span of currently experiencing violent conflict where cultural heritage is
two weeks, Quintus et al. (2017) evaluated LiDAR covering approxi- at risk.
mately 5–10 km2, whereas semi‐automated OBIA methods have • Second, we must continue developing new approaches that com-
allowed for the systematic evaluation of LiDAR datasets covering bine automated analysis with manual evaluation and subsequent
thousands of square kilometres in the same amount of time (e.g. Davis field‐testing to create a comprehensive landscape survey proce-
et al., 2018). dure. Each of these levels are essential for understanding the
One thing that is certain is that ground‐testing and manual archaeological record. By combining them together, we can study
analysis following automated detection algorithms is an essential landscape patterns at multiple scales, which is a vital component
step (Ainsworth, Oswald, & Went, 2013; Freeland et al., 2016, of landscape level archaeological research (e.g. Crumley, 1979;
72; Quintus, Clark, Day, & Schwert, 2015). Without ground‐ Millican, 2012; Robinson, 2010).
surveys, we cannot confirm the results of remotely sensed analysis
• Third, future work with OBIA should seek to compare different
and our knowledge cannot pass beyond a theoretical level.
methods of automated feature detection (e.g. Davis et al., in
Although in some instances automated survey may be inappropri-
review). By comparing different methods, researchers can best
ate, we must remember that ground‐survey, manual evaluation,
determine which methods are most appropriate for specific pur-
and automated detection algorithms are all useful tools for archae-
poses and thereby adopt the successes and avoid the failures
ologists, and all possess their own benefits and drawbacks. As
and setbacks of prior studies.
such, archaeologists must not exclude any of these methods out-
• Fourth, to improve the ability of OBIA to detect archaeological
right (sensu Hacιgüzeller, 2012).
features, researchers must share their datasets – this includes
new algorithms, computer code, processing steps, and training
4 | FUTURE DIRECTIONS data. By sharing this information, archaeologists around the world
can contribute to and access different methods and necessary
Where do we go from here? Are OBIA and automated object extrac- training data, thereby increasing and improving the use of OBIA
tion valuable methods for archaeologists? The answer is a resounding for archaeological problems. By making code and data available
yes. For areas at risk of development, destruction, or other distur- to all, even the non‐specialist can utilize some of these methods
bance, OBIA provides an efficient way to survey entire landscapes and contribute to the use of automated object detection.
with reasonable accuracy. As such, OBIA methods can serve in a
• Finally, archaeologists should use OBIA for studies beyond the
capacity to stop the development and destruction of areas containing
mere detection of features. Researchers can use detected
cultural deposits and better understand the spatial distribution of
objects to discuss broader spatial patterns of the archaeological
human settlements.
record (e.g. Freeland et al., 2016). Although the discovery of
Additionally, the use of OBIA allows for the re‐visitation of areas
new features is important, it is equally important to begin
and requires far less time and money than a pedestrian style ground
analysing this newly generated information to further our
survey (e.g. Bennett et al., 2014, 897; Davis et al., 2018). This is not
understanding of the human past.
to say that OBIA can detect everything there is to find in any given
area; not even manual evaluation can do that. Rather, OBIA can pro-
vide a baseline by which to assess the probability of cultural features
being present and set in motion a series of intensive ground surveys 5 | CO NC LUSIO NS
to validate these conclusions.
Even if researchers are reluctant to use these methods for the This article has sought to demonstrate the important advances that
detection of features, OBIA can still be useful in an archival sense. have occurred in applications of OBIA methods within landscape
Segmentation procedures can digitize archaeological deposits automat- archaeology. It has also traced some possible paths for the future of
ically with spatial and morphological accuracy. Archaeologists these methods within the discipline. A lot of progress has been made,
have done this with artefacts for statistical and morphometric analyses and yet there is still a great deal of untapped potential for OBIA to
(e.g. Aprile, Castellano, & Eramo, 2014; Hein, Rojas‐Domínguez, expand our understanding of the archaeological record. In the future,
Ornelas, D’Ercole, & Peloschek, 2018; Hofmann, Marschallinger, we should seek to incorporate (semi‐)automated algorithms with man-
Unterwurzacher, & Zobl, 2013; Lamotte & Masson, 2016) and have used ual analysis to ensure the broadest range of data is acquired. The
similar methods to study site formation processes (e.g. Sanger, 2015). importance of systematic documentation is vital in a world that suffers
There is a lot to gain from OBIA, especially in terms of understanding from cultural site destruction on a daily basis. OBIA is one method that
landscape‐level archaeological patterns. However, there are certain can help to record, preserve, protect, and study the record of our col-
avenues of research where these methods are yet to be fully invested: lective human history.

• First and foremost, the use of OBIA methods must be expanded

into new geographic areas where they have been under‐utilized ACKNOWLEDGEMENTS
or where they are yet to be introduced (e.g. North America, South The author would like to thank the anonymous peer reviewers for
America, Africa, coastal islands, etc., see Table 1). This is especially their helpful comments on this manuscript. Any and all errors are the
important for areas at risk of destruction from sea‐level rises or responsibility of the author.
DAVIS 7

FUND ING Cordero Ruiz, T., Cerrillo Cuenca, E., & Pereira, C. (2017). Detección de un
nuevo Campamento Romano en las inmediaciones de Mérida mediante
This research did not receive any specific grant from funding agencies tecnología LiDAR. SAGVNTVM. Papeles Del Laboratorio de Arqueología
in the public, commercial, or not‐for‐profit sectors. de Valencia, 49, 197. https://doi.org/10.7203/SAGVNTVM.49.10025
Crumley, C. L. (1979). Three locational models: An epistemological assess-
DECLARATION OF C ONFLICTS OF INTEREST ment for anthropology and archaeology. Advances in Archaeological
Method and Theory, 2, 141–173.
The author does not have any conflicts to declare, financial or Custer, J. F., Eveleigh, T., Klemas, V., & Wells, I. (1986). Application of
otherwise. Landsat data and synoptic remote sensing to predictive models for pre-
historic archaeological sites: An example from the Delaware Coastal
Plain. American Antiquity, 51(03), 572–588. https://doi.org/10.2307/
ORCID 281753
Dylan S. Davis https://orcid.org/0000-0002-5783-3578 Davis, D. S., Lipo, C. P., & Sanger, M. C. (in press). A comparison of auto-
mated object extraction methods for earthwork feature identification
RE FE R ENC E S in South Carolina. Journal of Archaeological Science: Reports, In Press.
https://doi.org/10.1016/j.jasrep.2018.10.035
Ainsworth, S., Oswald, A., & Went, D. (2013). Remotely acquired, not
remotely sensed: using lidar as a field survey tool. In R. S. Opitz, & D. Davis, D. S., Sanger, M. C., & Lipo, C. P. (2018). Automated mound detec-
C. Cowley (Eds.), Interpreting archaeological topography (pp. 206–222). tion using LiDAR survey in Beaufort County, SC. Southeastern
Oxford, UK: Oxbow Books. Archaeology, 1–15. https://doi.org/10.1080/0734578X.2018.1482186

Aprile, A., Castellano, G., & Eramo, G. (2014). Combining image analysis De Laet, V., Music, B., Paulissen, E., & Waelkens, M. (2008). Extracting
and modular neural networks for classification of mineral inclusions archaeological features from very high resolution QuickBird‐2 remote
and pores in archaeological potsherds. Journal of Archaeological Science, sensing imagery: A methodological approach based on the town of
50, 262–272. https://doi.org/10.1016/j.jas.2014.07.017 Sagalassos. In P. Degryse, & M. Waelkens (Eds.), Sagalassos VI: Geo‐
and bio‐archaeology at Sagalassos and in its territory (pp. 157–171). Leu-
Arvor, D., Durieux, L., Andrés, S., & Laporte, M.‐A. (2013). Advances in geo- ven, The Netherlands: Leuven University Press.
graphic object‐based image analysis with ontologies: A review of main
De Laet, V., Paulissen, E., Meuleman, K., & Waelkens, M. (2009). Effects of
contributions and limitations from a remote sensing perspective. ISPRS
image characteristics on the identification and extraction of archaeo-
Journal of Photogrammetry and Remote Sensing, 82, 125–137. https://
logical features from Ikonos‐2 and Quickbird‐2 imagery: Case study
doi.org/10.1016/j.isprsjprs.2013.05.003
Sagalassos (southwest Turkey). International Journal of Remote Sensing,
Beck, A., Philip, G., Abdulkarim, M., & Donoghue, D. (2007). Evaluation of 30(21), 5655–5668. https://doi.org/10.1080/01431160802705821
Corona and Ikonos high resolution satellite imagery for archaeological
De Laet, V., Paulissen, E., & Waelkens, M. (2007). Methods for the
prospection in western Syria. Antiquity, 81(311), 161–175. https://
extraction of archaeological features from very high‐resolution
doi.org/10.1017/S0003598X00094916
Ikonos‐2 remote sensing imagery, Hisar (southwest Turkey). Journal
Bennett, R., Cowley, D., & De Laet, V. (2014). The data explosion: tackling the of Archaeological Science, 34(5), 830–841. https://doi.org/10.1016/j.
taboo of automatic feature recognition in airborne survey data. Antiquity, jas.2006.09.013
88(341), 896–905. https://doi.org/10.1017/S0003598X00050766
D'Orazio, T., Palumbo, F., & Guaragnella, C. (2012). Archaeological trace
Bennett, R., Welham, K., Hill, R. A., & Ford, A. L. J. (2012). The application extraction by a local directional active contour approach. Pattern Recogni-
of vegetation indices for the prospection of archaeological features in tion, 45(9), 3427–3438. https://doi.org/10.1016/j.patcog.2012.03.003
grass‐dominated environments: Application of vegetation indices in
Drager, D. L. (1983). Projecting archaeological site concentrations in the
grass‐dominated environments. Archaeological Prospection, 19(3),
San Juan Basin, New Mexico. In D. L. Drager, & T. R. Lyons (Eds.),
209–218. https://doi.org/10.1002/arp.1429
Remote Sensing in Cultural Resource Management. Washington, DC:
Bescoby, D. J. (2006). Detecting Roman land boundaries in aerial photo- National Park Service.
graphs using Radon transforms. Journal of Archaeological Science,
Figorito, B., & Tarantino, E. (2014). Semi‐automatic detection of linear
33(5), 735–743. https://doi.org/10.1016/j.jas.2005.10.012
archaeological traces from orthorectified aerial images. International
Blaschke, T. (2010). Object based image analysis for remote sensing. ISPRS Journal of Applied Earth Observation and Geoinformation, 26, 458–463.
Journal of Photogrammetry and Remote Sensing, 65(1), 2–16. https://doi. https://doi.org/10.1016/j.jag.2013.04.005
org/10.1016/j.isprsjprs.2009.06.004
Freeland, T., Heung, B., Burley, D. V., Clark, G., & Knudby, A. (2016). Auto-
Blaschke, T., Hay, G. J., Kelly, M., Lang, S., Hofmann, P., Addink, E., … Tiede, mated feature extraction for prospection and analysis of monumental
D. (2014). Geographic Object‐Based Image Analysis – Towards a new earthworks from aerial LiDAR in the Kingdom of Tonga. Journal of
paradigm. ISPRS Journal of Photogrammetry and Remote Sensing, 87, Archaeological Science, 69, 64–74. https://doi.org/10.1016/j.
180–191. https://doi.org/10.1016/j.isprsjprs.2013.09.014 jas.2016.04.011
Campbell, K. M. (1981). Remote sensing: Conventional and infrared imag- Gheyle, W., Stichelbaut, B., Saey, T., Note, N., Van den Berghe, H., Van
ery for archaeologists. University of Calgary Archaeology Association, Eetvelde, V., … Bourgeois, J. (2018). Scratching the surface of war. Air-
11, 1–8. borne laser scans of the Great War conflict landscape in Flanders
Casana, J. (2014). Regional‐scale archaeological remote sensing in the age (Belgium). Applied Geography, 90, 55–68. https://doi.org/10.1016/j.
of big data. Advances in Archaeological Practice, 2(03), 222–233. apgeog.2017.11.011
https://doi.org/10.7183/2326‐3768.2.3.222 GRASS Development Team. (2018). Geographic Resources Analysis Sup-
Cerrillo‐Cuenca, E. (2017). An approach to the automatic surveying of pre- port System (GRASS) Software (Version 7.4). Open Source Geospatial
historic barrows through LiDAR. Quaternary International, 435, Foundation. Retrieved from https://grass.osgeo.org
135–145. https://doi.org/10.1016/j.quaint.2015.12.099 Guyot, A., Hubert‐Moy, L., & Lorho, T. (2018). Detecting neolithic burial
Chen, L., Comer, D. C., Priebe, C. E., Sussman, D., & Tilton, J. C. (2013). mounds from LiDAR‐derived elevation data using a multi‐scale
Refinement of a method for identifying probable archaeological sites approach and machine learning techniques. Remote Sensing, 10(2),
from remotely sensed data. In D. C. Comer, & M. J. Harrower (Eds.), 225. https://doi.org/10.3390/rs10020225
Mapping Archaeological Landscapes from Space ( ed., Vol. 5) (pp. Hacιgüzeller, P. (2012). GIS, critique, representation and beyond. Journal of
251–258). New York, NY: Springer. https://doi.org/10.1007/978‐1‐ Social Archaeology, 12(2), 245–263. https://doi.org/10.1177/
4614‐6074‐9_21 1469605312439139
8 DAVIS

Hanson, W. S. (2010). The future of aerial archaeology in Europe. Photo Lasaponara, R., & Masini, N. (2018). Space-Based Identification of Archae-
Interprétation. European Journal of Applied Remote Sensing, 46(1), 3–11. ological Illegal Excavations and a New Automatic Method for Looting
Harrower, M. J., Schuetter, J., McCorriston, J., Goel, P. K., & Senn, M. J. Feature Extraction in Desert Areas. Surveys in Geophysics, 39(6),
(2013). Survey, automated detection, and spatial distribution analysis 1323–1346. https://doi.org/10.1007/s10712‐018‐9480‐4
of Cairn Tombs in ancient southern Arabia. In D. C. Comer, & M. J. Har- Magnini, L., Bettineschi, C., & De Guio, A. (2017). Object‐based shell cra-
rower (Eds.), Mapping Archaeological Landscapes from Space ( ed., Vol. 5) ters classification from LiDAR‐derived sky‐view factor. Archaeological
(pp. 259–268). New York, NY: Springer. https://doi.org/10.1007/978‐ Prospection, 24(3), 211–223. https://doi.org/10.1002/arp.1565
1‐4614‐6074‐9_22
Menze, B. H., Mühl, S., & Sherratt, A. G. (2007). Virtual survey on north
Hay, G. J., & Castilla, G. (2008). Geographic Object‐Based Image Analysis Mesopotamian tell sites by means of satellite remote sensing. In B.
(GEOBIA): A new name for a new discipline. In T. Blaschke, S. Lang, & Ooghe, & G. Verhoeven (Eds.), Broadening Horizons: Multidisciplinary
G. Hay (Eds.), Object‐based Image Analysis (pp. 75–89). Heidelberg, Ger- approaches to landscape study (pp. 5–29). Newcastle, UK: Cambridge
many: Springer. https://doi.org/10.1007/978‐3‐540‐77058‐9_4 Scholars Publishing.
Hay, G. J., Castilla, G., Wulder, M. A., & Ruiz, J. R. (2005). An auto- Menze, B. H., & Ur, J. A. (2012). Mapping patterns of long‐term settlement
mated object‐based approach for the multiscale image in northern Mesopotamia at a large scale. Proceedings of the National
segmentation of forest scenes. International Journal of Applied Earth Academy of Sciences, 109(14), E778–E787. https://doi.org/10.1073/
Observation and Geoinformation, 7(4), 339–359. https://doi.org/ pnas.1115472109
10.1016/j.jag.2005.06.005
Menze, B. H., Ur, J. A., & Sherratt, A. G. (2006). Detection of ancient settle-
Hein, I., Rojas‐Domínguez, A., Ornelas, M., D'Ercole, G., & Peloschek, L. ment mounds: Archaeological survey based on the SRTM terrain
(2018). Automated classification of archaeological ceramic materials model. Photogrammetric Engineering & Remote Sensing, 72(3),
by means of texture measures. Journal of Archaeological Science: 321–327. https://doi.org/10.14358/PERS.72.3.321
Reports, 21, 921–928. https://doi.org/10.1016/j.jasrep.2017.12.032
Meredith‐Williams, M. G., Hausmann, N., Bailey, G. N., King, G. C. P.,
Hofmann, P., Marschallinger, R., Unterwurzacher, M., & Zobl, F. (2013).
Alsharekh, A., Al Ghamdi, S., & Inglis, R. H. (2014). Mapping, modelling
Marble provenance designation with object based image analysis:
and predicting prehistoric coastal archaeology in the southern Red Sea
state‐of‐the‐art rock fabric characterization from petrographic micro-
using new applications of digital‐imaging techniques. World Archaeol-
graphs. Austrian Journal of Earth Sciences, 106(2), 40–49.
ogy, 46(1), 10–24. https://doi.org/10.1080/00438243.2014.890913
InterIMAGE. (2009). InterIMAGE – Interpreting Images freely. Retrieved
Millican, K. (2012). The outside inside: Combining aerial photographs,
from http://www.lvc. ele.puc‐rio.br/projects/interimage/
cropmarks and landscape experience. Journal of Archaeological Method
Jahjah, M., Ulivieri, C., Invernizzi, A., & Parapetti, R. (2007). Archaeological and Theory, 19(4), 548–563. https://doi.org/10.1007/s10816‐012‐
remote sensing application pre‐post war situation of Babylon archaeo- 9140‐9
logical site—Iraq. Acta Astronautica, 61(1–6), 121–130. https://doi.org/
Mountrakis, G., Im, J., & Ogole, C. (2011). Support vector machines in remote
10.1016/j.actaastro.2007.01.034
sensing: A review. ISPRS Journal of Photogrammetry and Remote Sensing,
Jensen, J. R. (2007). Remote Sensing of the Environment: An earth resource 66(3), 247–259. https://doi.org/10.1016/j.isprsjprs.2010.11.001
perspective (2nd ed.). Upper Saddle River, NJ: Pearson Prentice Hall.
Ozawa, K. (1978). Classification of the keyhole shaped tombs by template
Kirk, S. D., Thompson, A. E., & Lippitt, C. D. (2016). Predictive modeling for matching method. IEEE Transactions on Computers, 27(5), 462–467.
site detection using remotely sensed phenological data. Advances in
Archaeological Practice, 4(1), 87–101. https://doi.org/10.7183/2326‐ Parcak, S. H. (2009). Satellite Remote Sensing for Archaeology. New York,
3768.4.1.87 NY: Routledge. https://doi.org/10.4324/9780203881460

Knoth, C., & Nüst, D. (2017). Reproducibility and practical adoption of Pregesbauer, M., Trinks, I., & Neubauer, W. (2014). An object oriented
GEOBIA with open‐source software in docker containers. Remote Sens- approach to automatic classification of archaeological features in mag-
ing, 9(3), 290. https://doi.org/10.3390/rs9030290 netic prospection data. Near Surface Geophysics, 12(2036). https://doi.
org/10.3997/1873‐0604.2014014
Körting, T. S., Garcia Fonseca, L. M., & Câmara, G. (2013). GeoDMA—geo-
graphic data mining analyst. Computers & Geosciences, 57, 133–145. Quintus, S., Clark, J. T., Day, S. S., & Schwert, D. P. (2015). Investigating
https://doi.org/10.1016/j.cageo.2013.02.007 regional patterning in archaeological remains by pairing extensive sur-
vey with a Lidar dataset: The case of the Manu’a Group, American
Kumar, M. J., Raj Kumar, G., & Reddy, V. K. (2014). Review on image seg-
Samoa. Journal of Archaeological Science: Reports, 2, 677–687. https://
mentation techniques. International Journal of Scientific Research
doi.org/10.1016/j.jasrep.2014.11.010
Engineering & Technology, 3(6), 992–997.
Kvamme, K. (2013). An examination of automated archaeological feature Quintus, S., Day, S. S., & Smith, N. J. (2017). The efficacy and analytical
recognition in remotely sensed imagery. In A. Bevan, & M. Lake importance of manual feature extraction using Lidar datasets. Advances
(Eds.), Computational Approaches to Archaeological Spaces (pp. 53–68). in Archaeological Practice, 5(4), 351–364. https://doi.org/10.1017/
Walnut Creek, CA: Left Coast Press. aap.2017.13

Lambers, K. (2018). Airborne and spaceborne remote sensing and digital Riley, M. A. (2009). Automated detection of prehistoric conical burial
image analysis in archaeology. In C. Siart, M. Forbriger, & O. Bubenzer mounds from LIDAR bare‐earth digital elevation models (Master's the-
(Eds.), Digital Geoarchaeology (pp. 109–122). Cham, Switzerland: sis). Department of Geology and Geography, Northwest Missouri State
Springer International Publishing. https://doi.org/10.1007/978‐3‐ University, Maryville, Missouri.
319‐25316‐9_7 Robinson, D. W. (2010). Land use, land ideology: An integrated geographic
Lamotte, A., & Masson, É. (2016). Arché‐OBIA: Un concept d’analyse quan- information systems analysis of rock art within south‐central California.
titative d’images numériques appliqué aux bifaces du gisement de American Antiquity, 75(04), 792–818. https://doi.org/10.7183/0002‐
Gouzeaucourt (Nord, FR). Notae Praehistoricae, 36, 121–130. 7316.75.4.792
Lasaponara, R., Leucci, G., Masini, N., & Persico, R. (2014). Investigating Sanger, M. C. (2015). Determining depositional events within shell deposits
archaeological looting using satellite images and GEORADAR: The using computer vision and photogrammetry. Journal of Archaeological
experience in Lambayeque in North Peru. Journal of Archaeological Sci- Science, 53, 482–491. https://doi.org/10.1016/j.jas.2014.10.026
ence, 42, 216–230. https://doi.org/10.1016/j.jas.2013.10.032 Schneider, A., Takla, M., Nicolay, A., Raab, A., & Raab, T. (2015). A
Lasaponara, R., & Masini, N. (2007). Detection of archaeological crop template‐matching approach combining morphometric variables for
marks by using satellite QuickBird multispectral imagery. Journal of automated mapping of charcoal kiln sites: Automated mapping of char-
Archaeological Science, 34(2), 214–221. https://doi.org/10.1016/j. coal kiln sites. Archaeological Prospection, 22(1), 45–62. https://doi.org/
jas.2006.04.014 10.1002/arp.1497
DAVIS 9

Schuetter, J., Goel, P., McCorriston, J., Park, J., Senn, M., & Harrower, M. Van Ess, M., Becker, H., Fassbinder, J., Kiefl, R., Lingenfelder, I., Schreier,
(2013). Autodetection of ancient Arabian tombs in high‐resolution G., & Zevenbergen, A. (2006). Detection of looting activities at archae-
satellite imagery. International Journal of Remote Sensing, 34(19), ological sites in Iraq using Ikonos imagery. In J. Strobl, T. Blaschke, & G.
6611–6635. https://doi.org/10.1080/01431161.2013.802054 Griesebner (Eds.), Angewandte Geoinformatik; Beiträge zum ( ed., Vol.
Sevara, C., Pregesbauer, M., Doneus, M., Verhoeven, G., & Trinks, I. (2016). 18) (pp. 668–678). Heidelberg, Germany: Wiechmann‐Verlag.
Pixel versus object – a comparison of strategies for the semi‐ Verhagen, P., & Drăguţ, L. (2012). Object‐based landform delineation and
automated mapping of archaeological features using airborne laser classification from DEMs for archaeological predictive mapping. Journal
scanning data. Journal of Archaeological Science: Reports, 5, 485–498. of Archaeological Science, 39(3), 698–703. https://doi.org/10.1016/j.
https://doi.org/10.1016/j.jasrep.2015.12.023 jas.2011.11.001
Sevara, C., & Pregesbauer, M. (2014). Archaeological feature classification: Wang, S., Hu, Q., Wang, F., Ai, M., & Zhong, R. (2017). A microtopographic
An object oriented approach. Southeastern European Journal of Earth feature analysis‐based LiDAR data processing approach for the identi-
Observation and Geomatics, 3(2), 139–143. fication of Chu tombs. Remote Sensing, 9(9), 880. https://doi.org/
Toumazet, J.‐P., Vautier, F., Roussel, E., & Dousteyssier, B. (2017). Auto- 10.3390/rs9090880
matic detection of complex archaeological grazing structures using Weng, Q. (2010). Remote Sensing and GIS Integration: Theories, methods, and
airborne laser scanning data. Journal of Archaeological Science: Reports, applications. New York, NY: McGraw‐Hill.
12, 569–579. https://doi.org/10.1016/j.jasrep.2017.03.012 Witharana, C., Ouimet, W. B., & Johnson, K. M. (2018). Using LiDAR and
Traviglia, A., & Torsello, A. (2017). Landscape pattern detection in archae- GEOBIA for automated extraction of eighteenth–late nineteenth cen-
ological remote sensing. Geosciences, 7(4), 128. https://doi.org/ tury relict charcoal hearths in southern New England. GIScience &
10.3390/geosciences7040128 Remote Sensing, 55(2), 183–204. https://doi.org/10.1080/
Trier, Ø. D., Larsen, S. Ø., & Solberg, R. (2009). Automatic detection of circular 15481603.2018.1431356
structures in high‐resolution satellite images of agricultural land. Zingman, I., Saupe, D., Penatti, O. A. B., & Lambers, K. (2016). Detection of
Archaeological Prospection, 16(1), 1–15. https://doi.org/10.1002/arp.339 fragmented rectangular enclosures in very high resolution remote
Trier, Ø. D., & Pilø, L. H. (2012). Automatic detection of pit structures in sensing images. IEEE Transactions on Geoscience and Remote Sensing,
airborne laser scanning data: Automatic detection of pits in ALS data. 54(8), 4580–4593. https://doi.org/10.1109/TGRS.2016.2545919
Archaeological Prospection, 19(2), 103–121. https://doi.org/10.1002/
arp.1421
Trier, Ø. D., Zortea, M., & Tonning, C. (2015). Automatic detection How to cite this article: Davis DS. Object‐based image analysis:
of mound structures in airborne laser scanning data. Journal of
a review of developments and future directions of automated
Archaeological Science: Reports, 2, 69–79. https://doi.org/10.1016/j.
jasrep.2015.01.005 feature detection in landscape archaeology. Archaeological
Trimble (2016). eCognition version 9.2 (p. 1). Germany: Trimble Germany Prospection. 2018;1–9. https://doi.org/10.1002/arp.1730
GmbH, Munich.