Publicação

Enhancing Density Current Analysis Through Machine Learning and Image Processing Techniques

Ver documento

Detalhes bibliográficos
Resumo:Laboratory measurements of density current propagation are usually carried out through image acquisition methods. Processing these images allows for the extraction of funda- mental flow parameters such as the current boundary, front velocity, head position, and density field. This processing generally requires complex computational routines, and the objective of this dissertation is to create more efficient and robust routines using the Python programming language. This dissertation investigates the application of Machine Learning (ML) techniques in image processing for the study of density currents. The specific objectives of this dissertation include developing Python codes for distortion correction and density field estimation, and integrating machine learning techniques to improve image processing. The development of the code for density field estimation was carried out by translating a MATLAB code used by Alves [5]. With the Python code for density field estimation developed, its improvement through the implementation of machine learning could be addressed. The density profiles obtained through the implementation of a Random Forest Regressor showed smoother transitions than the original, and were also less sensitive to noise. Additionally, model evaluation resulted in low Mean Squared Error (MSE) values. These observations demonstrate potential improvements in the accuracy and robustness of the density profile calculations through the integration of machine learning.
Autores principais:Martins, Emanuel Eugênio Silva
Assunto:Image Processing Density Currents Lock Exchange Experiment Machine Learning Python
Ano:2024
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL
Descrição
Resumo:Laboratory measurements of density current propagation are usually carried out through image acquisition methods. Processing these images allows for the extraction of funda- mental flow parameters such as the current boundary, front velocity, head position, and density field. This processing generally requires complex computational routines, and the objective of this dissertation is to create more efficient and robust routines using the Python programming language. This dissertation investigates the application of Machine Learning (ML) techniques in image processing for the study of density currents. The specific objectives of this dissertation include developing Python codes for distortion correction and density field estimation, and integrating machine learning techniques to improve image processing. The development of the code for density field estimation was carried out by translating a MATLAB code used by Alves [5]. With the Python code for density field estimation developed, its improvement through the implementation of machine learning could be addressed. The density profiles obtained through the implementation of a Random Forest Regressor showed smoother transitions than the original, and were also less sensitive to noise. Additionally, model evaluation resulted in low Mean Squared Error (MSE) values. These observations demonstrate potential improvements in the accuracy and robustness of the density profile calculations through the integration of machine learning.