Abstract
Plastic waste has been rapidly accumulating in the natural environment for decades sincethe commercial introduction of plastic in the 1950s as a disposable material. Millions of
tonnes of durable, persistent plastic waste now contaminate global ecosystems, including
rivers, urban beaches, remote islands, the deep sea, soils, mountains, and very small
pieces (microplastics) have even been detected in the air. Such pervasive plastic debris is
a deadly and common threat to wildlife due to ingestion and entanglement. Large debris
are particularly problematic not only because of these direct consequences, but because
they break down into smaller pieces in the environment. These smaller pieces are more
easily ingested by a greater diversity of species and are more difficult to remediate. It is
therefore sensible to focus prevention, remediation, and monitoring efforts on large, more
easily observable debris. In the past five years, monitoring efforts have increasingly
included the investigation of automated plastic detection via remote sensing technologies
and techniques. This approach has the potential to reduce observer bias (associated with
manual, visual surveys), and realise regular and widespread plastic pollution monitoring,
with the ultimate goal of achieving environmentally relevant reductions in plastic production
and pollution. This thesis contributes to these aims by i) investigating the performance of
novel remote sensing methods for plastic detection, ii) developing and implementing a new
plastic detection algorithm, and iii) evaluating for the first time, the impact of operational and
environmental variables on these specific plastic detection methods.
A large proportion of plastic litter detection investigations have implemented inexpensive
and common digital cameras that operate in the visible light spectrum. This approach is
often limited in capability/reliability by inherent ambiguity, resulting from two-dimensional
imagery of materials of unverifiable chemical composition. This thesis investigates the
detection of shoreline plastic litter with the aerial deployment of a specialised, hyperspectral
camera that is sensitive to shortwave infrared (SWIR) light and that measures signal from
more than ten times as many wavelengths as a conventional digital camera. The rationale
for pursuing this research is further detailed in Chapter 1, as well as thesis structure and
objectives. The chosen SWIR sensing technology was selected from numerous other
sensor technologies introduced in Chapter 2, for its specific relevance to detecting plastic
based on chemical composition. Chapter 3 details the functional principles of this
technology, which sees widespread use in commercial recycling facilities to successfully
sort large volumes of different types of plastic. The criteria for selecting the specific SWIR
camera (“OCI-F SWIR” from BaySpec, San Jose, CA, USA) implemented in this work, and an initial evaluation of its capabilities under a controlled lab environment are also presented
in Chapter 3, prior to assessing field performance in Chapters 4 and 5.
Chapter 4 presents the development of a customised, software-based plastic detection
algorithm, informed by results from Chapter 3 and by methods implemented in the literature.
Spectral Correlation Mapping (SCM) was selected for its computational efficiency and
documented improvement over detection with the commonly applied Spectral Angle
Mapping (upon which SCM is based). This method achieved successful classification of
plastic and non-plastic pixels within imagery of individual items of polyethylene (PE) and
polypropylene (PP) – the most common polymers produced globally – placed on a sandy
shoreline. The images were collected with a distance of approximately one metre between
the camera and the plastic items to measure a strong signal from the plastic and to inform
algorithm parameter selection before increasing image complexity and introducing
environmental variability in Chapter 5.
Imagery analysed in Chapter 5 was collected on different days to evaluate the impacts on
plastic detection performance of: plastic type, cloud cover, beach substrate, plastic’s
exposure to the environment (harvested from domestic and environmental sources), and
the altitude of the camera above the plastic. The camera was first deployed on a “highline”
apparatus and manually conveyed across a fixed transect, at a height of five metres above
a variety of plastic items arranged on the beach below. The purpose of this approach was
to ensure the collection of data under steady, regular movement, which might not be as
easily achieved with an uncrewed aerial vehicle (UAV). SCM was not able to consistently
classify plastic pixels in each transect as PE nor PP when SCM was applied in the same
way it had been in Chapter 4 (to images recorded at a one metre distance, of individual
plastic items). The mathematical factors contributing to this discrepancy were investigated
and modifications were subsequently implemented to the application of SCM, to compare
segmented spectra. An additional, new algorithm called Reflectance Range Analysis (RRA)
was also developed and evaluated. The development of RRA was based on differences
observed in the spectral characteristics of plastic items and substrate measured in the aerial
imagery, and on recommendations in the literature to develop/implement algorithms that
detect the intensity of plastic’s characteristic absorption. Where SCM classifies every image
pixel based on spectral similarity to a selected reference, RRA reveals spectral differences
between image pixels without requiring similarity to a reference.
Chapter 6 concludes this work by summarising the key findings and demonstrating the place
of this research in the evolution of plastic detection with SWIR remote sensing.
Recommendations are made here for augmenting plastic detection capabilities through
further study and improvements to survey and equipment design, as well as leveraging the
strengths of multiple remote sensing technologies. The diversity of current, newly available,
and upcoming sensor technologies offers numerous opportunities to measure, monitor, and
collectively address the problem of human-driven plastic pollution.
| Date of Award | 9 May 2023 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Sponsors | NERC |
| Supervisor | Finlo Cottier (Supervisor), Bhavani Narayanaswamy (Supervisor) & Benjamin Williamson (Supervisor) |