Identification of archaeological sites with LiDAR data using machine learning and feature extraction

• Autor(a) principal: Costa, Miguel Reis da Conceição
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/38724
• topic: Archaeology; LiDAR; Machine learning; Remote sensing; Random forest; Gradient; Entropy

This dissertation was developed within the project ODYSSEY, funded by the PORTUGAL2020 program. The project ODYSSEY aims to implement a platform of geographic information destined to archaeology and heritage experts. Identifying archaeological sites provides information that allows the consolidation of heritage information sources. The identification and detection of archaeological sites on the platform are expected to be automatic using Deep learning and Machine Learning techniques and data obtained from non-intrusive methods like LiDAR. The objective of this dissertation is to develop a method that can identify archaeological objects using Machine Learning and Feature Extraction. For this, it is required pre-processing the data provided, since the data is not ready to be used, being required to extract specific images of the target objects to build a dataset needed for training the classifier. There’s also a need to apply feature extraction in order to improve the data for training the classifier. The dataset used in the experimental results of this dissertation corresponds to LiDAR data obtained in a region around the city of Viana do Castelo, north of Portugal. This data contains 136 annotated archaeological sites called “mamoas”. After the pre-processing, several tests were executed, starting by evaluating the efficiency, while using different sizes of the bounding boxes. Several Machine Learning methods were evaluated, such as Random Forest, Adaptive Boosting, Bagging, Extra Trees, Gradient Boosting and Histogram of Gradient Boosting. It was also applied some texture filters on the images obtained from the data in order to extract features to use them on the classifier. It was obtained a combination that provided better results that are going to be used in future work. Afterwards, it was implemented an inference method using a sliding window where the objective is to identify objects of interest present in the area providing with the respective geographic coordinates.