Abstract:
The recent technological advancements in the field of geoinformatics has enabled faster surveying techniques that are emerging in the field of archaeology. However, carrying out such surveys in faster means while preserving the original accuracies and reliability which are the properties of traditional archaeological surveys is still challenging. The lower costs, least possible manpower, and a minimal disturbance to the archaeological sites during the survey are also among the general expectations of archaeological surveys. The drone technology, image processing software and cloud-based spatial platforms with analysis capabilities can combinedly assist for achieving the above objectives. This research developed a semi-automated archaeological object detection algorithm which can extract archaeological objects from drone images. The study area of this research was the Ramba Raja Maha Viharaya Archaeological Monastery site situated in Hambantota, Sri Lanka. A series of drone images were acquired using DJI Phantom 4 RTK drone and 20 MP, 1-inch CMOS sensor. The acquired images were processed using image processing functions and object detection and extraction algorithms written in Python language. The results and the accuracy verifications depict that the process of extracting archaeological ruins from the drone images was successful and in an acceptable accuracy. The confusion matrix returned in the model training was used to calculate the accuracy of the model since the raw accuracy is not very reliable when measuring the performance of a neural network. The performance indicators of the confusion matrix, that is the precision and recall were 61.7% and 95.5% respectively.
