Detection and tracking of barchan dunes using artificial intelligence

Author(s)

Esteban Andrés Cúnez Benalcázar

• Document type: Master's Dissertation
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Mecânica
• Defense date: 2024-03-05

Examining board members

Erick de Moraes Franklin; Arthur Vieira da Silva Oliveira; Renato Fuzaro Miotto

• Advisor: Erick de Moraes Franklin

Abstract

Barchan dunes are commonly found on the surface of planets such as Earth and Mars, playing an important role in the evolution of landscapes and ecosystems. Over the last few decades, imagery from remote sensing has been a valuable tool for investigating the morphodynamics of barchans. Still, their complex interactions and transformations make them sometimes difficult to detect. In this study, we use the capabilities of Artificial Intelligence (AI) and neural networks to develop an accurate and efficient model for the automatic detection and tracking of barchan in different environments (Aquatic, eolian, and martian). First, we collected a dataset of aquatic barchan dunes using an experimental setup with a rectangular cross-section, and Aeolian and Martian barchan dunes using the HiRISE website satellites. Subsequently, with the use of the CVAT platform, these images were labeled and classified using the classes: Barchan to detect dune regions and Not a barchan for shapes that do not correspond to a barchan. Thus, this data- set was used to train and evaluate the accuracy of our model, based on the YOLOv8 (You Only Look Once-YOLO) convolutional neural network for segmentation, which can detect barchan dunes with a confidence score (Cs) above 70%. Finally, we calculated the main parameters from the automatic detection of dunes such as dimensions, shapes, and other main morphology pro- perties. From this technique, the dataset obtained in this study can be used for further studies and applications related to dune detection in remote locations, providing a valuable resource for researchers. Overall, our research shows the potential of artificial intelligence and neural networks in the fields of Physics and Geoscience, and how they can be used to overcome the challenges of studying complex natural phenomena on Earth and other celestial bodies. Using a YOLO network to accurately identify and study dunes can improve our understanding of sur- face processes that are important to shape our landscape. We plan to continue investigating the possibilities of using Artificial Intelligence in other types of barchan dunes interactions and natural formations (AU)

• FAPESP's process: 21/11470-4 - Growth of three-dimensional dunes under turbulent flows
• Grantee: Esteban Andres Cuñez Benalcazar
• Support Opportunities: Scholarships in Brazil - Master