Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Sonarbildigenkänningssystem för att röja minor
KTH, School of Engineering Sciences (SCI), Mathematics (Dept.), Numerical Analysis, NA.
2019 (Swedish)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Sonar image recognition system to clear mines (English)
Abstract [sv]

Det finns ett intresse för att skapa säkra farvägar för militär och civil båttrafik. I svenska vatten ligger det kvar många havsminor från världskrigen som behöver röjas för att kunna skapa dessa vägar. Om en mina ska kunna röjas på ett säkert sätt är det till fördel att ha information om vilken bottentopografi som den befinner sig i. Detta görs genom att samla in sonarbilder ifrån autonoma undervattensfarkoster, AUV:er.

I denna rapport undersöks sonarbilderna genom att analysera en bit i taget med hjälp av olika kvadratiska filterstorlekar som roteras. Den bildbit som filtret täcker analyseras med hjälp applicering av en Gaussianfunktion som med hjälp av diskret Fourier Transform tar fram fyra olika varianser. Haar- funktionen har ej använts då den inte har samma noggrannhet i rumsdimensionen som Gaussian-funktionen har. Dessa varianser beskriver hur bildbiten varierar i en punkt, en linje, i radie samt i vinkel.

Olika kombinationer av dessa varianser leder till olika inneboende dimensioner i bildbitens mittersta pixel. De inneboende dimensionerna beskriver bottentopografierna på olika sätt.

Med hjälp av deep learning appliceras metoderna på ett träningsset som består av ett antal sonarbilder som plockats ut för vardera dimension. Med hjälp av Fourier-metoden beskriven ovan bestäms gränserna med ett konfidensintervall på 95%. De gränser som fanns längst ifrån 0 samt 1 var 0.22257 respektive 0.77708.

Gränserna applicerades i algoritmen som sedan tog fram numeriska värden för de inneboende dimensionerna. Utifrån dessa resultat blev det möjligt att presentera säkra farvägar genom grönmarkerade områden i sonarbilden. De filter som genererade det noggrannaste resultatet var de minsta filtren för alla rotationer.

Abstract [en]

There is an interest to make safe routes for military and civil boat traffic.

In the Swedish seas there exist a lot of mines from the world wars that need to be removed to create these highways. If a mine is going to be demined in a safe way it is in favor to have information about what kind of seafloor topography it is laying on. This is done by collecting sonar images from autonomous underwater vehicles, AUV:s.

In this report the sonar images are analyzed by investigating a smaller piece of the image. This is done by using different quadratic filter sizes which are rotating. The image piece which the filter is covering is analyzed by applying a Gaussian function that in combination with a discrete Fourier transform is producing four different variances. The Haar-function is not used since it has a lower accuracy in the dimension of the room than the Gaussian function has.

These variances describes how the image piece varies in a point, a line, radius and angle.

Different combinations of those variances are resulting in different intrinsic dimensions in the middle pixel of the image piece. The different intrinsic dimensions describes the seafloor topography in different ways.

The methods are applied on a training set with deep learning. The training set consists of sonar images which are chosen for every dimension. Applying the above mentioned Fourier-method it is possible to choose the boundaries with a confidence interval of 95%. Those boundaries that were most far away from 0 and 1 were 0.22257 respectively 0.77708.

These limits are applied in the algorithm that later on determines the numerical values for the intrinsic dimensions. It was possible to present safe highways in green in the sonar image from these results. Those filters that generated the highest accuracy were the smallest filters for all rotations.

Place, publisher, year, edition, pages
2019.
Series
TRITA-SCI-GRU ; 2019:054
National Category
Computational Mathematics
Identifiers
URN: urn:nbn:se:kth:diva-250231OAI: oai:DiVA.org:kth-250231DiVA, id: diva2:1307253
External cooperation
Försvarets Materielverk
Subject / course
Scientific Computing
Educational program
Master of Science - Applied and Computational Mathematics
Supervisors
Examiners
Available from: 2019-04-26 Created: 2019-04-26 Last updated: 2019-04-26Bibliographically approved

Open Access in DiVA

fulltext(1489 kB)15 downloads
File information
File name FULLTEXT01.pdfFile size 1489 kBChecksum SHA-512
4c93b10643969bae12656158b7ea91ce4e329d9b034c32119c22d267a680659b1982d40cef3a88d1f91d9c2634b08de8b57c86d3119ce6f1de663b6b9e5b0529
Type fulltextMimetype application/pdf

By organisation
Numerical Analysis, NA
Computational Mathematics

Search outside of DiVA

GoogleGoogle Scholar
Total: 15 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 47 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf