Universiteit Leiden

nl en

Conference | Digital Archaeology Group Meeting

Pre-CAA Digital Archaeology Group Special

12 March 2018
Digital Archaeology Group
Van Steenis Building
Einsteinweg 2
2333 CC Leiden

A fair few researchers at Leiden University are going to present at the next CAA conference in Tubingen. We are organising a DAG special session on the 12th of March in which these researchers will present their project here at the faculty. This gives them a chance to receive some feedback, as well as offering anyone who can't attend the CAA a chance to see some of the presentations.

Preliminary programme (more to follow!):

Reconstructing activities at a stratified site invariably includes a discussion of the constant diachronic change in the settlement. A case in point is the Late Bronze Age settlement at Tell Sabi Abyad, Syria. Here, a continuing reappropriation of space during the life span of the settlement presents a challenge for a functional analysis of the buildings and courtyards. Similarly, considering that the function of individual buildings was in constant flux, how can the settlement as a whole be characterized per phase or period?
In this paper, using 3D GIS and event-based chronologies for individual rooms, I will discuss a method for the analysis and presentation of the fluctuating use of space. The results aid in the advancement beyond simplistic settlement chronologies of distinct periods, towards an appreciation of overlapping timescales within the life time of a settlement.

Nowadays the surface of the earth is constantly being monitored by a multitude of airborne and satellite sensors that record a wide variety of environmental parameters. Over the last decade archaeologists have handled this ever-growing set of remotely sensed data by using computer-aided methods for the (semi-) automatic detection of archaeological objects. While successful, these handcrafted algorithms are highly specialized on specific object categories and data sources, which limits their use in different contexts. To overcome these limitations this research project will explore recent advancements in computer sciences in order to develop a generic, flexible and robust automated detection method for archaeological objects in remotely sensed data. Deep learning, a machine learning approach built on Convolutional Neural Networks (CNNs) seems especially promising.

In this paper a potential new technique for the automated detection of archaeological objects in airborne laser scanning data (ALS or LiDAR) will be presented. The technique is based on R-CNNs (Regions with CNN features). Unlike normal CNNs, that classify the entire input image, R-CNNs address the problem of object detection, which requires correctly localizing and classifying objects within a larger image. The proposed technique has been trained and tested on LiDAR data gathered from a forested area in the central part of the Netherlands. This area contains a multitude of archaeological objects, including (Prehistoric) barrows, Celtic fields and (Medieval) charcoal kilns. By implementing this new technique we will be able to develop a method to automatically detect and categorize these archaeological objects.

The use of semi-automatic methods to detect and extract archaeological objects in LIDAR data is providing outstanding results. For several large areas of the world, however, LIDAR data are not available, and satellite imagery is often the only source of information that archaeologists can use. Therefore, it is necessary to develop new automatisms to analyse satellite imagery.

This paper proposes a ruleset developed in eCognition to detect and extract complex burial monuments in Arabia from WorldView-2 satellite imagery.

The ruleset is composed of three parts. The first part identifies candidate objects as single pixels using an adaptive template matching algorithm. The second part uses the identified candidate pixels as seeds for a region growing segmentation which creates the borders of the objects. This process is suitable to draw any object automatically because it does not consider any assumption regarding its shape. The third part of the ruleset classifies the candidate objects using a combination of thresholds and a random tree classifier.

The adaptive template matching code has been trained using 130 burial monuments located across 3 km2. The same objects have been used to train the random tree classifier together with 200 negative samples. The ruleset was then applied on 100 km2 of terra incognita to test its efficiency. The results of this test are highly significant. This new method was able to detect and extract almost 80% of the burial monuments verified in the area, with an index of false positives equal to 30% of the total detections."

Over 60.000 Dutch archaeological research reports are available online, and this number is growing by around 4.000 a year. Much of this grey literature threatens to end up in a proverbial graveyard, unread and unknown. However, the information contained in these documents can be of immense value.

Currently it is only possible to search through the metadata of these documents, mainly via the Archis database and DANS repository. However, these metadata are often limited and sometimes inconsistent, and don't capture the `by-catch opportunity'; i.e. a single Bronze Age find within a large Medieval excavation

To effectively index these texts, Named Entity Recognition (NER) is needed to correctly identify and distinguish between archaeological concepts. Standard approaches to NER (in related fields), are insufficient to deal with the peculiarities of these concepts. 

Some research has already been done on NER in archaeological texts, e.g. in the ARIADNE & Open Boek projects, but these are not combined with full-text search, or tend to focus on limited entity types, and not the full breadth of archaeological concepts.

This paper will present the first phase of AGNES, in which machine learning is used to perform NER. The project is in cooperation with LIACS, who provide a computer cluster with high computing power, allowing for the use of more resource intensive techniques. The identified entities are combined with a full-text index to create an effective online search tool.