Detecting archaeological surface ceramics using remote sensing based on low-altitude images and weak learners

Abstract

Remote Sensing science has become increasingly valuable for archaeological research (Traviglia and Torsello, 2017). Drones and low-altitude systems provide a cost-effective way to document archaeological sites. In 2020, Domínguez-Rodrigo et al. highlighted the increasing use of advanced technologies such as machine learning and artificial intelligence. However, the success of remote sensing approaches largely depends on the methodology used during the research process. Sometimes, the methodology may be inadequate, making it difficult to assess the results accurately. Consequently, interpreting the results to align with the research objectives becomes challenging. Our study suggests that low-altitude remote sensing sensors and advanced image-processing techniques have immense potential to revolutionise field archaeological research. We discovered that detecting archaeological surface ceramics through drone imagery poses a challenge due to an 'imbalanced data distribution' issue, which leads to an accuracy paradox. Our study aimed to develop a new, robust methodology to optimise archaeological surface ceramic detection by blending AI methodologies for non-uniformly distributed classes.