Javascript must be enabled for the correct page display

Source Localization of Objects moving along different Paths using Neural Networks and Artificial Lateral Lines

Kramer, Stijn (2019) Source Localization of Objects moving along different Paths using Neural Networks and Artificial Lateral Lines. Bachelor's Thesis, Artificial Intelligence.

[img]
Preview
Text
AI_BA_2019_STIJNKRAMER.pdf

Download (3MB) | Preview
[img] Text
toestemming.pdf
Restricted to Registered users only

Download (120kB)

Abstract

Fish can detect objects underwater using their lateral line organ. This organ detects changes in the water flow relative to the body of the fish. Using the excitation patterns of these flow detecting organs, the location and velocity of the object can be inferred. We show in this thesis that these organs can be mimicked with fluid flow sensors called Artificial Lateral Lines (ALLs). Neural networks can be trained on data gathered by such sensors to solve the problem of localizing and determining the velocity of the object. We trained an Echo State Network (ESN), a Convolutional Neural Network (CNN) and a Time Delay Neural Network (TDNN) on the task of localizing moving objects in a simulated environment. None of the networks were able to get accurate estimations on the objects velocity. The ESN was not able to accurately localize the object with a mean error of 50% of the total grid size. The CNN and TDNN however, were able to get down to a mean error of within 10% of the total grid size. The TDNN also proved to be very robust against noise with little to no increase in error with added noise. The path an object follows also seemed to make little difference for localization and determining its velocity.

Item Type: Thesis (Bachelor's Thesis)
Supervisor:
Supervisor nameSupervisor E mail
Netten, S.M. vanS.M.van.Netten@rug.nl
Wolf, B.J.B.J.Wolf@rug.nl
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 20 Feb 2019
Last Modified: 04 Mar 2019 10:35
URI: http://fse.studenttheses.ub.rug.nl/id/eprint/19185

Actions (login required)

View Item View Item