This paper discusses how the use of AI (artificial intelligence) detected later prehistoric field... more This paper discusses how the use of AI (artificial intelligence) detected later prehistoric field systems provides a more reliable base for reconstructing palaeodemographic trends, using the Netherlands as a case study. Despite its long tradition of settlement excavations, models that could be used to reconstruct (changes in) prehistoric land use have been few and often relied on (insufficiently mapped) nodal data points such as settlements and barrows. We argue that prehistoric field systems of field plots beset on all sides by earthen banks—known as Celtic fields—are a more suitable (i.e. less nodal) proxy for reconstructing later prehistoric land use.
For four 32.25 km2 case study areas in different geogenetic regions of the Netherlands, prehistoric land use surface areas are modelled based on conventional methods and the results are compared to the results we obtained by using AI-assisted detection of prehistoric field systems. The nationally available LiDAR data were used for automated detection. Geotiff DTM images were fed into an object detection algorithm (based on the YOLOv4 framework and trained with known Dutch sites), and resultant geospatial vectors were imported into GIS.
Our analysis shows that AI-assisted detection of prehistoric embanked field systems on average leads to a factor 1.84 increase in known surface areas of Celtic fields. Modelling the numbers of occupants from this spatial coverage, yields population sizes of 37–135 persons for the case study regions (i.e. 1.15 to 4.19 p/km2). This range aligns well with previous estimates and offers a more robust and representative proxy for palaeodemographic reconstructions. Variations in land use coverage between the regions could be explained by differences in present-day land use and research intensity. Particularly the regionally different extent of forestlands and heathlands (ideal for the (a) preservation and (b) automated LiDAR detection of embanked field systems) explains minor variations between the four case study regions.
<p>The mining region of Upper Silesia has a long tradition with international signi... more <p>The mining region of Upper Silesia has a long tradition with international significance. In 2017, the historic silver mine in Tarnowsky Gory was recognized as a UNESCO World Heritage Site. With the mining of galena (PbS), the region developed into one of the most important industrial centers in Central Europe in the 16th century. In addition to the underground galleries, the historical mining has left thousands of mining shafts as small relief forms, which have not been systematically investigated so far. Partly the mining shafts are associated with Relict Charcoal Hearths (RCH), another small form which is a result of charcoal production. In the Mala Panew River valley, north of Tarnowsky Gory, several tens of thousands of these RCH are found, which could be mapped by LiDAR in recent years. More detailed pedological investigations, which would allow a systematic comparison with other known RCH sites, are missing so far.</p><p>Within the framework of a Polish-German cooperation project, we started in 2021 to investigate the mining shafts and the RCH in Tarnowsky Gory and in the Mala Panew River valley from a pedological-sedimentological point of view. At the RCH sites on the Mala Panew River, we focused on the following questions: How was the soil stratigraphy changed by the RCH construction? What are main processes of soil development before and after RCH construction? What was the role of the pits surrounding the RCH? How do the sites differ from the RCHs at Tarnowsky Gory especially with respect to soil properties and soil genesis? In Tarnowsky Gory, where a RCH was excavated directly next to a mining shaft, the following questions were in focus: How did the mining activity change soil distribution and soil properties? What are main processes of soil development on the different parts? What is the origin of the pit infill? What is the origin of the shaft rim deposits?</p><p>Our work program included the construction of excavator trenches across the mining remains, construction, description and sampling of soil profiles along the trenches, schematic drawing of the soil stratigraphy, and laboratory analyses for the determination of texture, Munsell color, pH (CaCl2, H20), CaCO3 content, Ctotal & Ntotal and total elements by FPXRF. We present the first results of the ongoing investigations.</p>
BoundingBox Localizer Tool (BLT), Geopodoly and Landscape Development department at the Brandenbu... more BoundingBox Localizer Tool (BLT), Geopodoly and Landscape Development department at the Brandenburgische Technische Universität Cottbus-Senftenberg version<br> Developed by W.B. Verschoof-van der Vaart MA & A. Brandsen MSc<br> Faculty of Archaeology / Data Science Research Programme<br> Leiden University, The Netherlands BLT transforms the output of an object detection model (such as Faster R-CNN) into geospatial vectors (polygons) usable in a GIS environment.
Although the history of automated archaeological object detection in remotely sensed data is shor... more Although the history of automated archaeological object detection in remotely sensed data is short, progress and emerging trends are evident. Among them, the shift from rule-based approaches towards machine learning methods is, at the moment, the cause for high expectations, even though basic problems, such as the lack of suitable archaeological training data are only beginning to be addressed. In a case study in the central Netherlands, we are currently developing novel methods for multi-class archaeological object detection in LiDAR data based on convolutional neural networks (CNNs). This research is embedded in a long-term investigation of the prehistoric landscape of our study region. We here present an innovative integrated workflow that combines machine learning approaches to automated object detection in remotely sensed data with a two-tier citizen science project that allows us to generate and validate detections of hitherto unknown archaeological objects, thereby contributi...
This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multi... more This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multiple archaeological object classes in LiDAR data from the Netherlands. WODAN2.0 is developed to rapidly and systematically map archaeology in large and complex datasets. To investigate its practical value, a large, random test dataset—next to a small, non-random dataset—was developed, which better represents the real-world situation of scarce archaeological objects in different types of complex terrain. To reduce the number of false positives caused by specific regions in the research area, a novel approach has been developed and implemented called Location-Based Ranking. Experiments show that WODAN2.0 has a performance of circa 70% for barrows and Celtic fields on the small, non-random testing dataset, while the performance on the large, random testing dataset is lower: circa 50% for barrows, circa 46% for Celtic fields, and circa 18% for charcoal kilns. The results show that the introdu...
This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multi... more This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multiple archaeological object classes in LiDAR data from the Netherlands. WODAN2.0 is developed to rapidly and systematically map archaeology in large and complex datasets. To investigate its practical value, a large, random test dataset—next to a small, non-random dataset—was developed, which better represents the real-world situation of scarce archaeological objects in different types of complex terrain. To reduce the number of false positives caused by specific regions in the research area, a novel approach has been developed and implemented called Location-Based Ranking. Experiments show that WODAN2.0 has a performance of circa 70% for barrows and Celtic fields on the small, non-random testing dataset, while the performance on the large, random testing dataset is lower: circa 50% for barrows, circa 46% for Celtic fields, and circa 18% for charcoal kilns. The results show that the introdu...
This paper discusses how the use of AI (artificial intelligence) detected later prehistoric field... more This paper discusses how the use of AI (artificial intelligence) detected later prehistoric field systems provides a more reliable base for reconstructing palaeodemographic trends, using the Netherlands as a case study. Despite its long tradition of settlement excavations, models that could be used to reconstruct (changes in) prehistoric land use have been few and often relied on (insufficiently mapped) nodal data points such as settlements and barrows. We argue that prehistoric field systems of field plots beset on all sides by earthen banks—known as Celtic fields—are a more suitable (i.e. less nodal) proxy for reconstructing later prehistoric land use.
For four 32.25 km2 case study areas in different geogenetic regions of the Netherlands, prehistoric land use surface areas are modelled based on conventional methods and the results are compared to the results we obtained by using AI-assisted detection of prehistoric field systems. The nationally available LiDAR data were used for automated detection. Geotiff DTM images were fed into an object detection algorithm (based on the YOLOv4 framework and trained with known Dutch sites), and resultant geospatial vectors were imported into GIS.
Our analysis shows that AI-assisted detection of prehistoric embanked field systems on average leads to a factor 1.84 increase in known surface areas of Celtic fields. Modelling the numbers of occupants from this spatial coverage, yields population sizes of 37–135 persons for the case study regions (i.e. 1.15 to 4.19 p/km2). This range aligns well with previous estimates and offers a more robust and representative proxy for palaeodemographic reconstructions. Variations in land use coverage between the regions could be explained by differences in present-day land use and research intensity. Particularly the regionally different extent of forestlands and heathlands (ideal for the (a) preservation and (b) automated LiDAR detection of embanked field systems) explains minor variations between the four case study regions.
&lt;p&gt;The mining region of Upper Silesia has a long tradition with international signi... more &lt;p&gt;The mining region of Upper Silesia has a long tradition with international significance. In 2017, the historic silver mine in Tarnowsky Gory was recognized as a UNESCO World Heritage Site. With the mining of galena (PbS), the region developed into one of the most important industrial centers in Central Europe in the 16th century. In addition to the underground galleries, the historical mining has left thousands of mining shafts as small relief forms, which have not been systematically investigated so far. Partly the mining shafts are associated with Relict Charcoal Hearths (RCH), another small form which is a result of charcoal production. In the Mala Panew River valley, north of Tarnowsky Gory, several tens of thousands of these RCH are found, which could be mapped by LiDAR in recent years. More detailed pedological investigations, which would allow a systematic comparison with other known RCH sites, are missing so far.&lt;/p&gt;&lt;p&gt;Within the framework of a Polish-German cooperation project, we started in 2021 to investigate the mining shafts and the RCH in Tarnowsky Gory and in the Mala Panew River valley from a pedological-sedimentological point of view. At the RCH sites on the Mala Panew River, we focused on the following questions: How was the soil stratigraphy changed by the RCH construction? What are main processes of soil development before and after RCH construction? What was the role of the pits surrounding the RCH? How do the sites differ from the RCHs at Tarnowsky Gory especially with respect to soil properties and soil genesis? In Tarnowsky Gory, where a RCH was excavated directly next to a mining shaft, the following questions were in focus: How did the mining activity change soil distribution and soil properties? What are main processes of soil development on the different parts? What is the origin of the pit infill? What is the origin of the shaft rim deposits?&lt;/p&gt;&lt;p&gt;Our work program included the construction of excavator trenches across the mining remains, construction, description and sampling of soil profiles along the trenches, schematic drawing of the soil stratigraphy, and laboratory analyses for the determination of texture, Munsell color, pH (CaCl2, H20), CaCO3 content, Ctotal &amp; Ntotal and total elements by FPXRF. We present the first results of the ongoing investigations.&lt;/p&gt;
BoundingBox Localizer Tool (BLT), Geopodoly and Landscape Development department at the Brandenbu... more BoundingBox Localizer Tool (BLT), Geopodoly and Landscape Development department at the Brandenburgische Technische Universität Cottbus-Senftenberg version<br> Developed by W.B. Verschoof-van der Vaart MA & A. Brandsen MSc<br> Faculty of Archaeology / Data Science Research Programme<br> Leiden University, The Netherlands BLT transforms the output of an object detection model (such as Faster R-CNN) into geospatial vectors (polygons) usable in a GIS environment.
Although the history of automated archaeological object detection in remotely sensed data is shor... more Although the history of automated archaeological object detection in remotely sensed data is short, progress and emerging trends are evident. Among them, the shift from rule-based approaches towards machine learning methods is, at the moment, the cause for high expectations, even though basic problems, such as the lack of suitable archaeological training data are only beginning to be addressed. In a case study in the central Netherlands, we are currently developing novel methods for multi-class archaeological object detection in LiDAR data based on convolutional neural networks (CNNs). This research is embedded in a long-term investigation of the prehistoric landscape of our study region. We here present an innovative integrated workflow that combines machine learning approaches to automated object detection in remotely sensed data with a two-tier citizen science project that allows us to generate and validate detections of hitherto unknown archaeological objects, thereby contributi...
This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multi... more This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multiple archaeological object classes in LiDAR data from the Netherlands. WODAN2.0 is developed to rapidly and systematically map archaeology in large and complex datasets. To investigate its practical value, a large, random test dataset—next to a small, non-random dataset—was developed, which better represents the real-world situation of scarce archaeological objects in different types of complex terrain. To reduce the number of false positives caused by specific regions in the research area, a novel approach has been developed and implemented called Location-Based Ranking. Experiments show that WODAN2.0 has a performance of circa 70% for barrows and Celtic fields on the small, non-random testing dataset, while the performance on the large, random testing dataset is lower: circa 50% for barrows, circa 46% for Celtic fields, and circa 18% for charcoal kilns. The results show that the introdu...
This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multi... more This paper presents WODAN2.0, a workflow using Deep Learning for the automated detection of multiple archaeological object classes in LiDAR data from the Netherlands. WODAN2.0 is developed to rapidly and systematically map archaeology in large and complex datasets. To investigate its practical value, a large, random test dataset—next to a small, non-random dataset—was developed, which better represents the real-world situation of scarce archaeological objects in different types of complex terrain. To reduce the number of false positives caused by specific regions in the research area, a novel approach has been developed and implemented called Location-Based Ranking. Experiments show that WODAN2.0 has a performance of circa 70% for barrows and Celtic fields on the small, non-random testing dataset, while the performance on the large, random testing dataset is lower: circa 50% for barrows, circa 46% for Celtic fields, and circa 18% for charcoal kilns. The results show that the introdu...
Nowadays archaeologists have vast amounts of LiDAR and other remote sensing data at their disposa... more Nowadays archaeologists have vast amounts of LiDAR and other remote sensing data at their disposal to search for previously unknown archaeological objects, often at a national scale. This leads to a Big Data problem in archaeology: some degree of automation is needed, as humans alone cannot cope with these ever-growing data sources. In this re-search, we have developed a novel workflow based on the Artificial Intelligence technology of Convolutional Neural Networks to automate the detection of unknown, complex archaeological objects. Our hypothesis is that a high-quality re-mote sensing data source such as LiDAR and a curated list of known objects is sufficient to find a large number — or ideally all — unknown other objects within a land-scape. In a case study presented here, we use Prehistoric hillforts in England as an example for this workflow and present a three-step approach to demonstrate the efficiency of the work-flow.
Uploads
For four 32.25 km2 case study areas in different geogenetic regions of the Netherlands, prehistoric land use surface areas are modelled based on conventional methods and the results are compared to the results we obtained by using AI-assisted detection of prehistoric field systems. The nationally available LiDAR data were used for automated detection. Geotiff DTM images were fed into an object detection algorithm (based on the YOLOv4 framework and trained with known Dutch sites), and resultant geospatial vectors were imported into GIS.
Our analysis shows that AI-assisted detection of prehistoric embanked field systems on average leads to a factor 1.84 increase in known surface areas of Celtic fields. Modelling the numbers of occupants from this spatial coverage, yields population sizes of 37–135 persons for the case study regions (i.e. 1.15 to 4.19 p/km2). This range aligns well with previous estimates and offers a more robust and representative proxy for palaeodemographic reconstructions. Variations in land use coverage between the regions could be explained by differences in present-day land use and research intensity. Particularly the regionally different extent of forestlands and heathlands (ideal for the (a) preservation and (b) automated LiDAR detection of embanked field systems) explains minor variations between the four case study regions.
For four 32.25 km2 case study areas in different geogenetic regions of the Netherlands, prehistoric land use surface areas are modelled based on conventional methods and the results are compared to the results we obtained by using AI-assisted detection of prehistoric field systems. The nationally available LiDAR data were used for automated detection. Geotiff DTM images were fed into an object detection algorithm (based on the YOLOv4 framework and trained with known Dutch sites), and resultant geospatial vectors were imported into GIS.
Our analysis shows that AI-assisted detection of prehistoric embanked field systems on average leads to a factor 1.84 increase in known surface areas of Celtic fields. Modelling the numbers of occupants from this spatial coverage, yields population sizes of 37–135 persons for the case study regions (i.e. 1.15 to 4.19 p/km2). This range aligns well with previous estimates and offers a more robust and representative proxy for palaeodemographic reconstructions. Variations in land use coverage between the regions could be explained by differences in present-day land use and research intensity. Particularly the regionally different extent of forestlands and heathlands (ideal for the (a) preservation and (b) automated LiDAR detection of embanked field systems) explains minor variations between the four case study regions.