MRAC Design for A Surveillance UAV for the Detection of Water Hyacinth
No Thumbnail Available
Date
2021-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Water hyacinth, locally named as ’Enboch’, is an invasive aquatic weed posing a
great threat to the worldwide aquatic ecosystem. Its existence has been reported to
greatly diminish water surfaces’ ecological value causing extensive nutrient reduction.
An intuitive, but much feasible and inexpensive solution relies on the early detection of
its presence followed by an action. This paper focuses on the design of a controller for a
quadrotor able to perform area surveillance specifically suited for the detection of the
hyacinth plant. The control design is done by taking the multivariate and non-linear
nature of the problem into full consideration. The developed model reference adaptive
controller (MRAC) comprising of both a standalone baseline controller and an adaptive
augmentation is found to be able to stabilize the system in nominal scenarios and also
restores nominal design performance in the presence of disturbances and parametric
uncertainties. For the task of water hyacinth detection, the technique of transfer
learning have been applied using the state-of-the-art VGG-16 model to perform feature
extraction for a CNN architecture. The problem has been formulated as a multi-class
classification problem considering three other aquatic plants identified as most probable
on the habitats of water hyacinth. The trained model obtained an accuracy level of
93.34% through the training phase, 94.25% on a validation set, and 93% on a testing set.
Description
Keywords
Water hyacinth, Model reference adaptive controller, Adaptive augmentation, Minimum snap trajectory, Transfer learning, Fine tuning, Convolutional neural network