Abstract
This article presents a novel deep learning method for semi-automated detection of historic mining pits using aerial LiDAR data. The recent emergence of national scale remotely sensed datasets has created the potential to greatly increase the rate of analysis and recording of cultural heritage sites. However, the time and resources required to process these datasets in traditional desktop surveys presents a near insurmountable challenge. The use of artificial intelligence to carry out preliminary processing of vast areas could enable experts to prioritize their prospection focus; however, success so far has been hindered by the lack of large training datasets in this field. This study develops an innovative transfer learning approach, utilizing a deep convolutional neural network initially trained on Lunar LiDAR datasets and reapplied here in an archaeological context. Recall rates of 80% and 83% were obtained on the 0.5 m and 0.25 m resolution datasets respectively, with false positive rates maintained below 20%. These results are state of the art and demonstrate that this model is an efficient, effective tool for semi-automated object detection for this type of archaeological objects. Further tests indicated strong potential for detection of other types of archaeological objects when trained accordingly.
Highlights
Airborne LiDAR systems are an increasingly valuable tool for locating, visualizing and understanding cultural heritage sites
In this paper we propose a highly effective transfer learning strategy to detect historic mining pits utilizing the DeepMoon base model fine-tuned with local LiDAR data
To fully test which LiDAR visualization is best suited for convolutional neural network (CNN) in future, would require a robust CNN trained from scratch on multiple differently visualized representations of the same data; such a model has not been made publicly available from any known sources at this time
Summary
Airborne LiDAR systems are an increasingly valuable tool for locating, visualizing and understanding cultural heritage sites. The English Heritage National Mapping Program (primarily aerial image interpretation) achieves a coverage rate of approximately 1 km per person per day; this project has been running for over 20 years employing on average 15–20 staff and had covered an area of 52,000 km by 2012, in contrast, the Baden-Württemberg study, whilst still a primarily manual approach, took advantage of automated processing where possible, allowing an estimated coverage rate of over 35,000 km by a single operator in six years [7] These two projects are not directly comparable, as the quality and accuracy of their results varies greatly along with the data types analyzed [3], but it is an indication of the speed advantages gained from integrating automated processes into an analysis workflow. With current advances in computing power the potential to pre-process entire national datasets in weeks rather than decades is a distinct possibility
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
