Object detection of megalithic dolmens in Google Satellite imagery

• Autor(a) principal: Marçal, Daniel André Barbosa
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/29930
• topic: Deep learning; Machine learning; Geology; Análise da imagem -- Image analysis; Visão computacional -- Computer vision; Geologia

The detection of dolmens, ancient megalithic structures, holds significant human and archaeological importance. We address the challenge of automating dolmen detection in the region of Alentejo, Portugal, by leveraging deep learning algorithms, specifically YOLOv8 and FasterRCNN, and exploring the potential influence of terrain information, including distance to water and rocky outcrops for refining the detection results. The motivation behind this research stems from the abundance of undiscovered dolmens in the area, prompting the need for an efficient and accurate detection pipeline. The methodology involves preprocessing the images to enhance relevant features and applying the Deep Learning approach for dolmen detection. The key finding of the study demonstrates the efficacy of FasterRCNN architectures in dolmen detection, which have achieved a confidence degree of 93% (average precision). These findings offer valuable insights and practical assistance to local archaeologists in identifying small megalithic structures in similar regions. However, limitations were encountered, mostly due to the unavailability of high-quality images in Google Earth databases, thereby affecting the precision of the results. Future work should focus on acquiring more comprehensive and high-resolution image datasets to enhance the performance of the detection algorithm. Our research provides a promising pipeline for automated dolmen detection using Deep Learning algorithms. However, we must emphasize the need for continued improvement and for the acquisition of additional data from different information sources to enhance the accuracy, efficiency, and generalization capacity of dolmen detection algorithms.