3oulder detection |
DOI: 10.23784/HN119-01
Automatic detection of boulders
by neural networks
A comparison of multibeam echo sounder
and side-scan sonar performance
An article by PETER FELDENS, PATRICK WESTFELD, JENNIFER VALERIUS, AGATA FELDENS and SVENJA PAPENMEIER
Neural networks show great promise in the automatic detection of boulders on the
seafloor. Maps derived from bathymetric data show better performance compared to
Dackscatter mosaics in this study. However, we find the lack of training data ground-
truthed to a high standard the largest challenge for automated object detection based
an acoustic data.
boulder detection | neural networks | hydrographic surveying | bathymetry | backscatter
Zrkennung von Felsbrocken | neuronale Netze | Seevermessung | Bathymetrie | Backscatter
Neuronale Netze sind sehr vielversprechend bei der automatischen Erkennung von Felsbrocken auf dem
Meeresboden. Aus bathymetrischen Daten abgeleitete Karten zeigen in dieser Studie eine bessere Leis-
zung im Vergleich zu Rückstreumosaiken. Die größte Herausforderung für die automatische Objekterken
nung auf Basis akustischer Daten ist jedoch der Mangel an Trainingsdaten, die auf einem hohen Standard
arprobt sind.
Authors
Dr. Peter Feldens, Agata
°eldens and Dr. Svenja
)apenmeier work at the
„eibniz Institute for Baltic Sea
kesearch Warnemünde.
Dr. Patrick Westfeld and
Jennifer Valerius work at
che Federal Maritime anc
Aydrographic Agency (BSH) in
'ostock and Hamburg
peter feldens@io-warnemuende.de
1 Introduction
Multibeam echo sounders (MBES) have been used
for decades to provide high-quality bathymetric
Maps of the seafloor (Lurton 2002; Augustin et al
‚996; Pickrill and Todd 2003). The German Hydro-
graphic Office (Federal Maritime and Hydrograph
ic Agency, BSH) collects bathymetry and detects
abjects underwater by vessel-mounted MBES
zystems (Dehling and Ellmer 2012). The data sur-
veyed in German waters are processed into official
nautical charts and nautical publications to ensure
navigational safety at sea. Accurate and reliable in
‘ormation of seabed’s topography further forms a
decisive basis for political and technical decisions
‚elating to the sea, including applications depend
ng On spatio-temporal-resolved 3-D geodata.
=cho sounding is a measurement technique al
'owing for the 3-D reconstruction of the surface of
:he seafloor and all objects located on it. As a pri
Mary result, a digital surface model (DSM) is avail
able. During the following data processing chain
zonducted at BSH, the task is to separate between
:he surface measured and the actual seabed, to
derive a digital terrain model (DTM) of the seafloor
"he detection and extraction of boulders are chal
jenging. At BSH, it is realised in a semi-automatic
process based on geometric filtering, with interac
tive post-processing and a final visual inspection
oy well-trained experts. This procedure is time-
“onsuming and error-prone because of subjectiv
ıty and generalisation. Against a background of
increasing user requirements (e.g. nautical infor-
mMation service needs a consistent separation of
seabed and boulders for chart production; marine
spatial planning requires information about condi-
tions on the seabed to assess the impact of off
shore construction projects) and the compliance
with international standards (IHO 5-44 Order 1a
and 1b require the reliable detection of obstacles
along all main shipping routes), automation of the
processing chain is indispensable in terms of ac
zuracy, reliability and reproducibility of the results
't is also required in the sense of an efficient evalu-
ation of large areas.
Next to hydrographic applications, recent de
velopments in habitat mapping require the de
‚ection of cobbles and larger hard substrates. The
'dentification of marine hard substrates based on
acoustic remote sensing is important for the de-
‚ection, delineation and ecological assessment of
seafloor habitats (Papenmeier et al. 2020) as wel
as for marine spatial planning. This need is ac
zounted for in several international frameworks,
such as the Convention on Biological Diversity
and the Marine Strategy framework directives.
Boulder detection in the German Baltic and the
North Sea for these purposes is done using side
scan sonar (SSS) systems. Next to the ease of op
eration over large scales, the survey geometry
of a side-scan sonar, towed above the seafloor.
Hydroaraphische Nachrichten
