What lies beneath? Machine learning and imaging of lichen-covered surfaces at Stonehenge to reveal rock art

Lead Research Organisation: University of Brighton
Department Name: Sch of Environment and Technology

Abstract

A quarter of the stone surface of Stonehenge is inaccessible to conventional archaeological survey techniques such as laser scanning due to the dense coverage of lichen. This project will aim to: (i) create a novel method, combining machine learning and surface imaging techniques such as photogrammetry, to reveal carvings that may be hidden by lichen covered stone surfaces; and (ii) verify any findings using a subsurface imaging technique such as terahertz imaging. The project will answer the question of whether there are more prehistoric rock carvings hidden by lichen on Stonehenge, and whether, with only an understanding of the topography of the lichen, we can find these carvings.

The results will be of use for further unravelling archaeological detail at Stonehenge, and aid conservation, presentation and management of the site. Outcomes will have wider applicability as a rapid non-invasive technique for measuring and monitoring the microtopography of vegetation-covered stone surfaces at other monuments and historic buildings.

Planned Impact

1. Academic beneficiaries: The CDT will develop scientific and engineering excellence in the domain of cultural heritage scientific and engineering research and more fundamentally in the enabling domains of imaging and sensing, visualisation, modelling, computational analysis and digital technology. While the CDT focusses on the complex materials and environments of the arts, heritage and archaeology, it will be broadly influential due to the range of novel methods and approaches to be developed in collaboration with the Diamond Light Source and the National Physical Laboratory. The establishment of a student and alumni-managed 'Heritage Science Research Network', will enable CDT's cross-disciplinarity to bridge EPSRC subject boundaries impacting scholarly research in the arts and humanities and social sciences.
2. Heritage beneficiaries: The CDT will have a transformational effect on public heritage institutions by dovetailing 'Data creation', 'Data to knowledge' and 'Knowledge to enterprise' research strands. The resulting advances in understanding, interpretation, conservation, presentation, management, communication, visualisation of heritage, and improved visitor participation and engagement will lead to significantly improved public service and value creation in this sector. This will sustainably boost the cultural heritage tourism sector which requires significant heritage science capacity to maintain the UK's cultural assets, i.e. museum, library, archive and gallery collections and historic buildings. 15 globally leading heritage Partner institutions (both national and international) will contribute to dissemination through established and new heritage networks e.g. the EU Heritage Portal (http://www.heritageportal.eu/).
3. Industry, particularly three crucial sectors: (i) sensors and instrumentation, which underpin a wide range of industrial activity despite the small size (UK Sales £3Bn), and are a key enabling technology for successful economic growth: 70% of the revenues of FTSE 100 companies (sales of £120Bn) are in sectors that are highly dependent on instrumentation; (ii) creative industries, increasingly vital to the UK with 2M employees in creative jobs and the sector contributing £60Bn a year (7.3%) to the UK economy. Over the past decade, the creative sector has grown at twice the rate of the economy as a whole; (iii) heritage tourism sector contributing £7.4Bn p.a. to the UK economy and supporting 466,000 equivalent jobs. Without the CDT, this crucially important economy sector will experience an unsustainable loss of capacity. The impact will be achieved in collaboration with our Partners: Electronics, Sensors, Photonics KTN, TIGA and Qi3, a technology commercialisation, business development and knowledge transfer company.
4. Public: The intensive public engagement activities are built into CDT including dissemination and engagement events at heritage institutions, popular science conferences and fora, e.g. Cheltenham Science Festival, European Science Open Forum and British Science Festival, as well as events organised by the HEIs' Beacon projects (e.g. UCL Bright Club). Cross-cohort encouragement to engage in these events will realise the substantial potential for the CDT to popularise science and engineering. More widely, visitors and users of heritage will benefit from the development of new and more engaging presentation tools, and pervasive and mobile computing.
5. Policy: SEAHA will engage with policy makers, by contributing evidence to policies and research agendas (the PI is actively involved in the EU JPI Cultural Heritage and Global Change, in which she advised on the development of the EU Cultural Heritage Research Agenda endorsed on 22/03/2013) and develop policy briefings for governmental and parliamentary bodies. The CDT is also a strategically important development of the AHRC/EPSRC Science and Heritage Programme ensuring continued global UK leadership in the SEAHA domain.

Publications

10 25 50
 
Description 1. Created a novel approach for visualising prehistoric carvings on stone surfaces obscured by lichen. The work is a proof-of-principle study to show that some crustose and foliose lichens growing on the stones of Stonehenge do not present enough noise to obscure carvings. The visualisation method uses a custom algorithm that highlights areas of high local curvature on the rock surface, which is indicative of carvings. It suggests that the methods used previously to identify the carvings at Stonehenge, namely laser scanning, may not be of high enough resolution to find lichen-obscured carvings. Or, existing imaging analysis workflows were of insufficient sensitivity to carvings beneath lichen. The resulting lichen-on-rock visualisation was however very noisy. An important research question is whether there are noise removal methods, particularly in machine learning, that could tackle the challenge this presents to identifying very eroded carvings beneath lichen.

2. Developed a methodology for using supervised learning to detect the presence of axe-head carvings on the stones of Stonehenge. A neural network for classifying 3-D object meshes called MeshNet was trained to differentiate between carving and non-carving photography-derived meshes. The carving meshes represented Early Bronze Age carvings identified on the stones of Stonehenge, while the non-carving meshes depicted generally flat, featureless surfaces of the stones. The highest classification accuracy achieved was 84.2% using 75 carvings and 19 non-carvings for training and testing the neural network respectively. MeshNet can be adapted to become an automatic carving recognition tool and be deployed alongside conventional carving visualisation techniques to verify manually found carvings and identify the location of unseen carvings. A research question this opens is whether the same methodology can be applied on carvings obscured by lichen to locate carvings beneath lichen using conventional photography.

3. Completed three short literature reviews on machine learning methods capable of revealing occlusions in visual data, procedures taken to mitigate the effect of lichens on rock art and contemporary subsurface imaging methods. Through the literature reviews, a supervised machine learning approach using point cloud data has been identified as appropriate for revealing occluded 3-D objects; due to the uncertainty of whether lichen protects or attacks its substrate, lichen removal has been identified as an inappropriate response to lichens growing on rock carvings; and terahertz imaging has been identified as a possible subsurface imaging method for revealing rock carvings hidden by lichen.

4. Another research question that has opened in the project is whether contemporary laser scanning techniques could be used to reveal rock carvings beneath the sparse but shrubby Ramalina siliquosa lichen populations prevalent on the stones of Stonehenge. An approach that uses high resolution, multi-perspective laser scanning combined with lichen removal post-processing is currently being explored.
Exploitation Route My research outcomes can be taken forward via an academic route by archaeologists and rock art conservators as a technique to accurately observe, record, and preserve rock art that was previously hidden by lichen. The combination of photogrammetry and machine learning will result in an easy to implement methodology that may only require taking pictures of the subject with a DSLR camera, followed by processing on a computer. The technique may not even require an expert in rock art to implement, only someone with experience in photogrammetry. The same technique can also aid in the detection of shallow and heavily eroded prehistoric carvings on non-lichenised rock surfaces. Through a non-academic route, the resulting visualisations can be used in the Stonehenge visitor centre and beyond, in an educational setting, to reveal the prehistoric cultural past of Britain. Curators, artists and teachers could use the 3-D data generated in this work to show in precise detail how the rock art was formed, with what implements and give a better understanding of who were the people that shaped Stonehenge and its environs.
Sectors Digital/Communication/Information Technologies (including Software),Environment,Culture, Heritage, Museums and Collections