Machine Learning and Image-Processing-Based Method for the Detection of Archaeological Structures in Areas with Large Amounts of Vegetation Using Satellite Images

Abstract

The detection of archaeological structures in satellite images is beneficial for archaeologists since it allows quick identification of structures across large areas of land. To date, some methods have been proposed to solve this task; however, these methods do not give good results in areas with large amounts of vegetation, such as those found in the southeast of Mexico and Guatemala. The method proposed in this paper works on satellite images obtained with SASPlanet. It uses two color spaces (RGB and HSL) and filters (Canny, Sobel, and Laplacian) jointly with supervised machine learning to improve the detection of archaeological structures in areas with a lot of vegetation. The method obtains an average performance of at least 93% on precision, recall, F1 score, and accuracy. Thus, our proposal is a very good option compared with traditional techniques for manual or semi-automatic detection of structures, identifying archaeological sites in a shorter time.