lnu.sePublikationer
RefereraExporteraLänk till posten
Permanent länk

Direktlänk
Referera
Referensformat
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Annat format
Fler format
Språk
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Annat språk
Fler språk
Utmatningsformat
  • html
  • text
  • asciidoc
  • rtf
Compare and evaluate unsupervised anomaly based and supervised classification methods for 2D wood defect detection
Linnéuniversitetet, Fakulteten för teknik (FTK), Institutionen för datavetenskap och medieteknik (DM).
Linnéuniversitetet, Fakulteten för teknik (FTK), Institutionen för datavetenskap och medieteknik (DM).
2026 (Engelska)Självständigt arbete på grundnivå (kandidatexamen), 10 poäng / 15 hpStudentuppsats (Examensarbete)
Hållbar utveckling
Berör inget SDG
Abstract [en]

Automated visual inspection is an important area within industrial manufacturing. Machine learning and computer vision are increasingly used for defect detection. This thesis investigates automated defect detection for treated wooden surfacesin an industrial environment. The motivation for the project is the need for a quality inspection system from an industrial window manufacturing company. Manualinspection is time consuming and subject to inconsistency, which motivates an investigation into a camera based machine learning system that could provide a moreconsistent and cost effective alternative. Existing research has mainly evaluated supervised and unsupervised methods under controlled conditions. This thesis willinstead evaluate the methods under identical production-like conditions on treatedwood surfaces. To address the problem, a prototype vision inspection system wasdeveloped using a 2D camera together with deep learning models for defect detection. A supervised classification model based on ResNet-18 and an unsupervisedanomaly detection model based on PatchCore were implemented and evaluated using both controlled and production-like image data. The results showed that bothmodels achieved relatively strong performance under controlled conditions. However, the production-like evaluation introduced substantially lower precision due to ahigh number of false positive predictions. The anomaly model demonstrated highersensitivity towards deviations and the classification model produced more selectivepredictions. A combined approach reduced false positives while maintaining a reasonable defect detection capability. The results show that combining supervised classification and unsupervised anomaly detection can provide a more balanced inspection strategy, although further development is required to make it a reliable inspectionsystem.

Ort, förlag, år, upplaga, sidor
2026. , s. 25
Nyckelord [en]
Industrial wood inspection, Supervised classification, Anomaly detection, PatchCore, ResNet-18
Nationell ämneskategori
Datavetenskap (datalogi)
Identifikatorer
URN: urn:nbn:se:lnu:diva-149746OAI: oai:DiVA.org:lnu-149746DiVA, id: diva2:2104491
Externt samarbete
Elitfönster AB
Ämne / kurs
Datavetenskap
Utbildningsprogram
Programvaruteknik, 180 hp
Presentation
2026-06-02, Homeros V, Universitetsplatsen 1, Växjö, 13:30 (Engelska)
Handledare
Examinatorer
Tillgänglig från: 2026-09-24 Skapad: 2026-09-23 Senast uppdaterad: 2026-09-24Bibliografiskt granskad

Open Access i DiVA

fulltext(3092 kB)71 nedladdningar
Filinformation
Filnamn FULLTEXT01.pdfFilstorlek 3092 kBChecksumma SHA-512
130ff5dedd1801e4696a0fad1b927a9b3978c3462fc2571eb31485bbff2f3570d1ec0853b8a08ac1d433c99d6de4f314db897f994de931a485e1d30fa6bad724
Typ fulltextMimetyp application/pdf

Sök vidare i DiVA

Av författaren/redaktören
Brisendal, JohannesArfvidsson, Max
Av organisationen
Institutionen för datavetenskap och medieteknik (DM)
Datavetenskap (datalogi)

Sök vidare utanför DiVA

GoogleGoogle Scholar
Totalt: 71 nedladdningar
Antalet nedladdningar är summan av nedladdningar för alla fulltexter. Det kan inkludera t.ex tidigare versioner som nu inte längre är tillgängliga.

urn-nbn

Altmetricpoäng

urn-nbn
Totalt: 4038 träffar
RefereraExporteraLänk till posten
Permanent länk

Direktlänk
Referera
Referensformat
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Annat format
Fler format
Språk
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Annat språk
Fler språk
Utmatningsformat
  • html
  • text
  • asciidoc
  • rtf