Analysis of background noise and boats noise detection in coastal areas

Abstract (Swedish)

The underwater noise has a significant negative impact on marine animals. The
EU has made efforts to keep noise emission within sustainable limits, but only
in regards to boats with Automatic Identification System (AIS). The noise of
recreational boats, which are the main noise emitter in coastal areas, is overlooked.
Thus, there is a need to quantify the noise emitted by recreational boats and, in
the extension, develop models for this noise. This thesis aims to give an overview
of how background noise varies at some coastal locations and to quantify the
contributions of boat noise based on recordings. In addition, the thesis aims to
design an energy detector, evaluate the detector applied to recorded noise, and
briefly outline alternative methods to detect boat noise. For background noise and
boat noise contribution purposes, a probabilistic power spectral density (PPSD) is
used. It is applied to a long window to see the background noise, and to two short
windows to be able to see the additional noise that is due to boat passages. The
background noise is successfully visualized, and the main contributing factor to
similar sound signatures in the PPSDs to visualize background noise in different
locations is the geometry of the seabed. A minimum level of boat noise contribution
can be found using this method, but it requires the recordings to not include
precipitation. The energy detector consisted of a rolling window structure with
a short and a long window, to compensate for the shifting background noise.
The designed detector could detect about half of the boat passages, but could be
improved by only considering a certain frequency range and adapting the threshold
to the application. Some alternative methods are to investigate the periodicity in
the tones of the boat noise or tonal detection. As a tonal detection algorithm, the
Estimation of Signal Parameters via Rotational Invariance Techniques (ESPRIT)
algorithm is implemented, but found not to be successful as a detector without the
use of Akaike Information Criterion (AIC) or Minimum Description Length (MDL)
to set a critical parameter. ESPRIT combined with one of the other algorithms is
deemed worthy of further investigation.

Popular Abstract

Studies have shown that the underwater noise undoubtedly has a negative impact
on marine animals. The EU have made efforts to keep the noise emission within
sustainable limits, but mainly in regards to commercial freight ships. This means
that noise from recreational boats, which are the main noise emitter in coastal
areas, are overlooked. The noise from recreational boats is also under-investigated,
and there are no estimates of the amount of this type of noise. Therefore, it
is important to study the noise generated by recreational boats. By doing so,
we can develop models to predict this noise and better assess its impact on the
environment.
This thesis aims to give an overview of how the background noise varies at coastal
locations and to quantify the boat noise contributions based on underwater recordings. Further, the thesis aims to construct and evaluate a method to detect boat
noise in the recordings and briefly outline alternative methods to detect boat noise.
Boat noise consists of broadband noise, for instance caused by propeller cavitation
and vibration from the hull, and narrowband tones consisting of a fundamental
tone and overtones, with a certain relation between the frequencies of the tones.
This leaves three main alternatives when trying to detect boat noise: Utilizing the
energy, the tonal components or the frequency relationship between the tones in
the boat noise.
A commonly used method of the first alternative is the energy detector, which
was used in the thesis. Through some modifications the detector would flag events
of increased energy relative to the temporary background energy. Though, the
method possessed a difficulty of handling precipitation of varying intensity. Overall, the detector identified about half of the boats, with only a few false detections.
However, it could be adjusted based on the specific terms of application.
As an alternative, a classical tonal detection algorithm called ESPRIT (Estimation of signal parameters via rotational invariance techniques) was implemented,
and recommended for further investigation, since this method does not possess
the weaknesses that the energy detector has, such as the difficulties in case of precipitation. The ESPRIT algorithm identifies the frequencies of the most probable
iii
tones within the noise. It always detects at least one tone, but can not determine
the exact number of tones. Thus, in combination with another algorithm that
determines if there are more than one tone present in the noise, the conclusion
should be that there is a boat present if there is more than one tone.
For the purpose of identifying the background noise and boat noise contribution,
another method, a probabilistic method was used. The method calculated the occurrence of different types of noise, or more in detail the combinations of frequency
and intensity, and visualized the more common noise, or background noise. The
main contributing factors of similar background noise at different locations was
the geometry of the seabed. This probabilistic method also quantifies the minimum levels of boat noise, assuming that there are no other noise sources, such as
precipitation.

• author: Lennartsson, Sophia
• supervisor: Fredrik Tufvesson
• organization: Department of Electrical and Information Technology
• course: EITM01 20241
• year: 2024
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• report number: LU/LTH-EIT 2024-1009
• language: English
• id: 9167517
• date added to LUP: 2024-08-15 13:05:03
• date last changed: 2024-08-15 13:05:03