PosterPDF Available

Performance of Neural Networks in Source Localization using Artificial Lateral Line Sensor Configurations

Authors:

Abstract

Artificial lateral lines (ALL) are used to detect the movement and locations of sources underwater, and are based on the lateral line organ found in fish and amphibians. Experiments have been performed to evaluate if the localization performance of neural networks, trained on simulated ALL sensor data, can be improved through adjustments of the internal ALL sensor positions. A Cramér-Rao lower bound analysis was performed on a subset of handpicked sensor configurations to estimate the likely performance of various configurations. The best and worst configurations were used to generate simulated datasets with which extreme learning machines (ELMs) and convolutional neural networks (CNNs) were trained and tested on their location accuracy. Simulated datasets consisted of two sources in a three-dimensional basin and the sensor readings of 16 ALL sensors. Results show that the best performing configuration consists of improved ELM and CNN localization performances, while also demonstrating that ELMs are capable of localizing multiple sources in three-dimensional aquatic environments, with comparable if not better results than CNNs.
P. van der Meulen, B.J. Wolf, P. Pirih, S.M. van Netten
University of Groningen
PPerformance of Neural Networks in Source Localization
using Artificial Lateral Line Sensor Configurat
ions
The optimal configuration improved performance for both sources,
compared to other configurations. Theref
ore, the main research
question can be answered positively in that using an optimal
configuration can imp
rove source localization performance using
CNNs.
With regard to the secondary research question, both neur
al
networks are capable of detecting two sources in a 3D environment,
if sources are an equal distance r
emoved from the ALL. If not, only
the closest source to the array is accurately reconstructed.
The optima
l configuration also improved ELM results for all source
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.
Artificial lateral lines (ALLs) are used to detect the movement and
locations of sources underwater,
and are based on the neuromasts
(fig.1) located in the lateral line organ found in fish and
amphibians. ALLs con
sists of a set of biaxial sensors (fig. 2)
Görner, P. (1963). Untersuchungen zur morphologie und elektrophysiologie des
seitenlinienorgans vom krallenfros
ch (xenopus laevis daudin). Journal of Comparative
Physiology A: Neuroethology, Sensory, Neural, and Beh
avioral Physiology, 47 (3), 316
338.
Wolf, B. J., Morton, J. A., MacPherson, W. B. N., & van Netten, S. M. (2
018). Bioinspired all-
optical artificial neuromast for 2d flow sensing. Bioinspiration & biomimetics.
Data generation:
source locations:
sensor locations:
teacher object:
3D matrix containing 1331 density
probability points for source locations
Neural networks:
convolutional neural network
extreme learning machine
sensor readings

3D matrix
Source prediction process:
3D matrix
source predictions
Can the placement of artificial lateral line sensors be beneficial for
improving the accuracy of source
localization through the use of
convolutional neural networks?
Are convolutional neural networks and ex
treme learning machines
capable of predicting the locations of multiple sources in three-
dimensional envi
ronments?
Fig. 1: Superficial neuromasts of a
clawed frog. From Görner (1963).
Fig. 2: Biaxial ALL sensor. From
Wolf
et al. (2018).
EXPERIMENT 1
A Crar-Rao lower bound analysis was performed on a subset of
sensor configurations (16 sensor
s, 1m3
basin) to estimate their
likely performances and indicate the best and worst configurations.
EXPERIMENT 2
The best and worst configurations were used to generate simulated
datasets to train and test extreme lear
ning machines (ELMs) and
convolutional neural networks (CNNs) on their location accuracy.
Simulated dataset
s consisted of 2 sources in a 3D basin (1m3
) and
the sensor readings of 16 ALL sensors.
Source localization pipeline
Calculate sensor readings
EXPERIMENT 1:
EXPERIMENT 2:
INTRODUCTION RESULTS
RESEARCH QUESTIONS
SOURCE LOCALIZATION PIPELINE
METHODS
CONCLUSION
REFERENCES
k-means
(b)
(a)
(a)(b)
Fig. 4: Bar plots for the normalized worst predicted source location and
prediction distribut
ions versus the distance between both source
locations (horizontal ALL, using
30:HorFours
). (a): CNN; (b): ELM. Colors
indicate distributions. With CNNs, the secondary source performance
shows a lin
ear relationship with the distance between sources, which is
not the case with ELMs.
Fig. 5: Estimated prob
ability curves for the distribution of total source-
prediction distances for the best (a) and worst (b
) predicted sources.
ALLs were placed horizontally at z:0. Curves were averaged over 5
repetitions and
4 dataset conditions.
Fig. 3: 2D sensor configurations for 16 sensors. Configurations were
applied hor
izontally (at z:0) and vertically (at y:0). The Crar-Rao
lower bound analysis indicates that
30:HorFours
(green) and
22:VertLineMid
(blue) performed best horizontally and vertically,
respectively.
5:HorLineLow
(red) is indicated as the worst performing
configuration.
Poster presented at the ICT.OPEN2018 conference, 19-20 March 2018, Amersfoort, Netherlands
... This type of network was also used in [14] to localize both moving and stationary vibrating sources in a 2D plane. Recently [24], the ELM architecture was compared with a recurrent network architecture (LSTM) for objects moving in a straight line in a 2D plane. To the authors' knowledge, the present work is the first effective demonstration of a CNN architecture for localizing a source with an ALL. ...
... Perhaps in a more complex setting it might still be valuable to incorporate previous time frames, as has been shown for localizing a single source using regression [24]. One such situation that may benefit from taking history into account is when objects turn more slowly than the current maximal 1 rad (57 degrees) per second or when the objects can vary their speed. ...
... The application of hydrodynamic detection of objects can therefore not scale up indefinitely, but the range can be extended beyond the current chosen domain. It has been shown that the source can be positioned in an area next to the array and still be detected reliably using different types of artificial neural networks [14,20,24]. It is, therefore, not necessary for the whole velocity pattern to be sampled within the area directly in front of the array; the domain can extend beyond the length of the array. ...
Article
Full-text available
This research focuses on the signal processing required for a sensory system that can simultaneously localize multiple moving underwater objects in a three-dimensional (3D) volume by simulating the hydrodynamic flow caused by these objects. We propose a method for localization in a simulated setting based on an established hydrodynamic theory founded in fish lateral line organ research. Fish neurally concatenate the information of multiple sensors to localize sources. Similarly, we use the sampled fluid velocity via two parallel lateral lines to perform source localization in three dimensions in two steps. Using a convolutional neural network, we first estimate a two-dimensional image of the probability of a present source. Then we determine the position of each source, via an automated iterative 3D-aware algorithm. We study various neural network architectural designs and different ways of presenting the input to the neural network; multi-level amplified inputs and merged convolutional streams are shown to improve the imaging performance. Results show that the combined system can exhibit adequate 3D localization of multiple sources.
ResearchGate has not been able to resolve any references for this publication.