University of Groningen
Performance of Neural Networks in Source Localization using Artificial Lateral Line Sensor
Configurations
van der Meulen, Pim; Wolf, Berend; Pirih, Primoz; van Netten, Sietse
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from
it. Please check the document version below.
Document Version
Final author's version (accepted by publisher, after peer review)
Publication date:
2018
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
van der Meulen, P., Wolf, B., Pirih, P., & van Netten, S. (2018). Performance of Neural Networks in Source
Localization using Artificial Lateral Line Sensor Configurations. Poster session presented at ICT OPEN
2018: The Interface for Dutch ICT-Research, Amersfoort, Netherlands.
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
P
.
van
de
r
Me
ul
e
n,
B.
J
.
Wol
f
,
P
.
Pi
r
i
h,
S.
M.
van
Ne
t
t
e
n
Uni
ve
r
s
i
t
y
of
Gr
oni
nge
n
P
Per
f
or
mance
of
Neur
al
Net
wor
ks
i
n
Sour
ce
Local
i
zat
i
on
usi
ng
Ar
t
i
f
i
ci
al
Lat
er
al
Li
ne
Sensor
Conf
i
gur
at
i
ons
Theoptimalconfiguration improved performanceforboth sources, compared to other configurations. Therefore, the main research question can be answered positively in that using an optimal configuration can improvesourcelocalization performanceusing CNNs.
With regard to the secondary research question, both neural networksarecapableofdetecting two sourcesin a 3D environment, ifsourcesarean equaldistanceremoved from theALL.Ifnot,only theclosestsourceto thearray isaccurately reconstructed.
Theoptimalconfiguration also improved ELM resultsforallsource generation conditions; the use of an ELM leads to a higher performance of the worst estimated source, for the majority of conditions,compared to using a CNN.
Artificiallaterallines(ALLs)areused to detectthemovementand locationsofsourcesunderwater,and arebased on theneuromasts (fig.1) located in the lateral line organ found in fish and amphibians.ALLsconsistsofa setofbiaxialsensors(fig.2)
Görner, P. (1963). Untersuchungen zur morphologie und elektrophysiologie des seitenlinienorgansvom krallenfrosch (xenopuslaevisdaudin).JournalofComparative Physiology A:Neuroethology,Sensory,Neural,and BehavioralPhysiology,47 (3),316– 338.
Wolf,B.J.,Morton,J.A.,MacPherson,W.B.N.,& van Netten,S.M.(2018).Bioinspired all -opticalartificialneuromastfor2d flow sensing.Bioinspiration & biomimetics.
Data generation: source locations: sensorlocations:
teacherobject: 3D matrix containing 1331 density probability pointsforsourcelocations Neuralnetworks:
convolutionalneuralnetwork extreme learning machine sensorreadings 3D matrix Sourceprediction process:
3D matrix source predictions
Can theplacementofartificiallaterallinesensorsbebeneficialfor improving theaccuracy ofsourcelocalization through theuseof convolutionalneuralnetworks?
Areconvolutionalneuralnetworksand extremelearning machines capableofpredicting thelocationsofmultiplesourcesin three -dimensionalenvironments?
Fig.1:Superficialneuromastsofa clawed frog.From Görner(1963).
Fig.2:BiaxialALL sensor.From Wolfetal.(2018).
EXPERIMENT 1
A Cramér-Rao lowerbound analysiswasperformed on a subsetof sensor configurations (16 sensors, 1m3 basin) to estimate their
likely performancesand indicatethebestand worstconfigurations.
EXPERIMENT 2
Thebestand worstconfigurationswereused to generatesimulated datasetsto train and testextremelearning machines(ELMs)and convolutionalneuralnetworks(CNNs)on theirlocation accuracy. Simulated datasetsconsisted of2 sourcesin a 3D basin (1m3)and
thesensorreadingsof16 ALL sensors. S o u rc e l o c a li z a ti o n p ip e li n e Calculatesensorreadings EXPERIMENT 1: EXPERIMENT 2:
I
NTRODUCTI
ON
RESULTS
RESEARCH
QUESTI
ONS
SOURCE
LOCALI
ZATI
ON
PI
PELI
NE
METHODS
CONCLUSI
ON
REFERENCES k-means (b) (a) (a) (b)Fig.4:Barplotsforthenormalized worstpredicted sourcelocation and prediction distributions versus the distance between both source locations(horizontalALL,using 30:HorFours).(a):CNN;(b):ELM.Colors indicatedistributions.With CNNs,thesecondary sourceperformance showsa linearrelationship with thedistancebetween sources,which is notthecasewith ELMs.
Fig.5:Estimated probability curvesforthedistribution oftotalsource -prediction distancesforthebest(a)and worst(b)predicted sources. ALLs were placed horizontally at z:0. Curves were averaged over 5 repetitionsand 4 datasetconditions.
Fig.3:2D sensorconfigurationsfor16 sensors.Configurationswere applied horizontally (at z:0) and vertically (at y:0).The Cramér-Rao lower bound analysis indicates that30:HorFours (green) and
22:VertLineMid (blue) performed best horizontally and vertically, respectively.5:HorLineLow(red)isindicated astheworstperforming configuration.