Theses and Dissertations
ORCID
https://orcid.org/0009-0006-0817-2626
Advisor
Samiappan, Sathishkumar
Committee Member
Turnage, Gray
Committee Member
Zope, Anup
Date of Degree
12-13-2024
Original embargo terms
Visible MSU Only 1 year
Document Type
Graduate Thesis - Campus Access Only
Major
Computational Engineering
Degree Name
Master of Science (M.S.)
College
James Worth Bagley College of Engineering
Department
Computational Engineering Program
Abstract
High-resolution imagery from Uncrewed Aerial Systems (UAS) equipped with multispec tral(MSI) and RGB sensors can be used for mapping invasive aquatic species. This research employs classification models like CNN, AlexNet, VGG16, ResNet, and DenseNet, alongside segmentation models such as FCN, U-Net, SegNet, DeepLabV3+, and Mask R-CNN, to analyze and visualize four invasive aquatic species including Cuban Bulrush (Oxycaryum cubense), Water Hyacinth (Eichhornia crassipes), Juncus (Juncus effusus), and Cattail (Typha latifolia). Among the tested models, ResNet (97.44% for RGB)andDenseNet(96.20%forMSI)exhibitedthehighest classification accuracies. For segmentation tasks, FCN (75.5% for RGB) and SegNet (79.6% for MSI) excelled. The study also examines the computational complexity and energy consumption of these models by linking energy use to carbon emissions given that 60% of global energy comes from fossil fuels. By measuring the carbon footprint of model training, this research emphasizes the ecological costs of deep learning technologies in environmental mapping.
Recommended Citation
Prakash, Ruchitha Yadav, "Mapping invasive aquatic plants using deep learning classification and segmentation models" (2024). Theses and Dissertations. 6602.
https://scholarsjunction.msstate.edu/td/6602
