Machine learning based approaches for classification of oil spills and microplastics in marine environments

Chen, Yifu (2021) Machine learning based approaches for classification of oil spills and microplastics in marine environments. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (2MB)

Abstract

Environmental modelling is an important approach of environmental engineering and management since it helps gain better understanding of environmental problems and impacts and facilitate environmental decision-making processes. However, because of the intricate conditions enormous data, diverse uncertainties, and various standards and requirements, environmental modeling is usually sophisticated and challenging. This study aimed to develop the novel modelling approaches by integrating machine learning (ML) into analyzing tabular and image datasets for environmental applications. Firstly, a data-driven binary classification approach was developed to analyze oil fingerprinting. After comparing six different machine learning algorithms on five different biomarkers, random forest classifier was found as the most effective and accurate model to distinguish weathered chemically dispersed and non-dispersed oil from the dataset of diamantanes. The developed model was approved to be capable of aiding oil fingerprinting under the studied conditions. It showed the good value of ML methods in environmental modeling especially for oil spill response research and practice. Secondly, an integrated approach by combing the strengths of convolutional neural networks and improved deep convolutional generative adversarial networks was proposed to classify microplastics and oil-dispersant agglomerates (MODAs) with diverse weathering conditions. The f score and model accuracy suggested the robust prediction from the trained model on the dataset of MODAs with different weathering degrees. The results could provide a better understanding of microplastics’ effects on oil fate and transport during a marine oil spill. The proposed approach also presented the high potential of facilitating image-related classification work in environmental fields. This dissertation not only developed two new ML based modelling approaches for environmental applications in oil fingerprinting and oil/microplastics classification, but also demonstrated the high value of ML methods and deep neural networks in processing experimental data for supporting environmental engineering and management.

Actions (login required)

View Item