@MASTERSTHESIS{pgi2020002,
	author = "L. Smith",
	supervisor = "M. Roper",
	title = "Detecting Coastal Litter with Neural Networks",
	school = "Department of Computer and Information Sciences, University of Strathclyde",
	year = "2019",
	abstract = "Coastal litter is a signi{\"\i}{\neg}…ant problem in Scotland, not only polluting and endangering
wildlife but also harming the vital tourism sector. Aerial images of the coastlines, as
well as spreadsheets containing information for the majority of these images, were made
available by Scrapbook, an organisation dedicated to combating this issue. The aim of
this dissertation is to utilise this information in the design, application and evaluation
of deep learning based systems for automatically classifying these aerial photographs by
their level of litter accumulations, and to form recommendations for a fully automated
system based on these results.
In the process, a variety of solutions to domain problems are explored, such as:
insu{\"\i}{\neg}ツient samples, class imbalance, massive terrain variety, and the automated col
lection of features for both incorporation to training and the automatic presentation
of {\"\i}{\neg}］dings. Additionally, signi{\"\i}{\neg}…ant processing of the supplied dataset was necessary,
some utilities involved forming a part of the proposed automated system.
Ultimately the limited dataset prevented the development of a su{\"\i}{\neg}ツiently e{\"\i}{\neg}�ective
model, performances poor for all classes other than the most numerous. However, the
e{\"\i}{\neg}ツacy of proposed methods such as data augmentation and mixed input networks was
validated, and it is proposed that through methodologies employed in this dissertation,
when more images become available, an accurate system can be deployed",
}