Exploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS data

In this master thesis deep learning is proven to be applicable in the field of seasurveillance. Commercial ships using the AIS system have to report the type of thevessel such as fishing ship or cargo ship. A problem with AIS data is that it can beeasily manipulated and therefore deliberately or acc...

Full description

Bibliographic Details
Main Author: Ljunggren, Henrik
Format: Others
Language:English
Published: KTH, Mekatronik 2017
Subjects:
Online Access:http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-217526
id ndltd-UPSALLA1-oai-DiVA.org-kth-217526
record_format oai_dc
spelling ndltd-UPSALLA1-oai-DiVA.org-kth-2175262020-01-29T03:40:27ZExploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS dataengEn studie över tillämpligheten av deep learning inom sjöövervakningLjunggren, HenrikKTH, Mekatronik2017Engineering and TechnologyTeknik och teknologierIn this master thesis deep learning is proven to be applicable in the field of seasurveillance. Commercial ships using the AIS system have to report the type of thevessel such as fishing ship or cargo ship. A problem with AIS data is that it can beeasily manipulated and therefore deliberately or accidentally incorrect. This thesis will focus on detecting false ship types. To detect a false ship type 19 different methods were tested on the 1100 hour long AIS data set. Three of these methods were baseline methods using a more conventional approach to the sea surveillanceproblem. The testing showed that the best performing method was one of the deeplearning methods proving that deep learning is indeed suitable in sea surveillance. Detta examensarbetet bevisar att neurala nätverk är tillämpningsbara inom sjöövervakning. Kommersiella skepp som använder Automatic Identification System(AIS) måste rapportera sin skeppstyp exempelvis fiskebåt eller transportbåt. Ett problem med AIS data är att den är lätt att manipulera och kan därför medvetet eller omedvetet vara felaktig. Detta examensarbete fokuserar på att upptäcka falska skeppstyper genom att analysera båtrörelser. För att upptäcka falska skeppstyper har 19 olika metoder testats på en 1100 timmar lång AIS datamängd. Tre av dessa metoder var standardmetoder som var uppbyggda genom mer konventionella tillvägagångssätt.Testerna visade att den bäst lämpade metoden för sjöövervakning varen av deep learning metoderna. Student thesisinfo:eu-repo/semantics/bachelorThesistexthttp://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-217526TRITA-ITM-EX ; 2017:153application/pdfinfo:eu-repo/semantics/openAccess
collection NDLTD
language English
format Others
sources NDLTD
topic Engineering and Technology
Teknik och teknologier
spellingShingle Engineering and Technology
Teknik och teknologier
Ljunggren, Henrik
Exploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS data
description In this master thesis deep learning is proven to be applicable in the field of seasurveillance. Commercial ships using the AIS system have to report the type of thevessel such as fishing ship or cargo ship. A problem with AIS data is that it can beeasily manipulated and therefore deliberately or accidentally incorrect. This thesis will focus on detecting false ship types. To detect a false ship type 19 different methods were tested on the 1100 hour long AIS data set. Three of these methods were baseline methods using a more conventional approach to the sea surveillanceproblem. The testing showed that the best performing method was one of the deeplearning methods proving that deep learning is indeed suitable in sea surveillance. === Detta examensarbetet bevisar att neurala nätverk är tillämpningsbara inom sjöövervakning. Kommersiella skepp som använder Automatic Identification System(AIS) måste rapportera sin skeppstyp exempelvis fiskebåt eller transportbåt. Ett problem med AIS data är att den är lätt att manipulera och kan därför medvetet eller omedvetet vara felaktig. Detta examensarbete fokuserar på att upptäcka falska skeppstyper genom att analysera båtrörelser. För att upptäcka falska skeppstyper har 19 olika metoder testats på en 1100 timmar lång AIS datamängd. Tre av dessa metoder var standardmetoder som var uppbyggda genom mer konventionella tillvägagångssätt.Testerna visade att den bäst lämpade metoden för sjöövervakning varen av deep learning metoderna.
author Ljunggren, Henrik
author_facet Ljunggren, Henrik
author_sort Ljunggren, Henrik
title Exploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS data
title_short Exploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS data
title_full Exploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS data
title_fullStr Exploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS data
title_full_unstemmed Exploring the capabilities of deep learning in seasurveillance : Using deep learning to classify motion trajectories from AIS data
title_sort exploring the capabilities of deep learning in seasurveillance : using deep learning to classify motion trajectories from ais data
publisher KTH, Mekatronik
publishDate 2017
url http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-217526
work_keys_str_mv AT ljunggrenhenrik exploringthecapabilitiesofdeeplearninginseasurveillanceusingdeeplearningtoclassifymotiontrajectoriesfromaisdata
AT ljunggrenhenrik enstudieovertillamplighetenavdeeplearninginomsjoovervakning
_version_ 1719310151524024320