• norsk
    • English
  • English 
    • norsk
    • English
  • Login
View Item 
  •   Home
  • Fakultet for ingeniørvitenskap (IV)
  • Institutt for maskinteknikk og produksjon
  • View Item
  •   Home
  • Fakultet for ingeniørvitenskap (IV)
  • Institutt for maskinteknikk og produksjon
  • View Item
JavaScript is disabled for your browser. Some features of this site may not work without it.

Computer Vision for Defect Detection in Wood Manufacturing

Neidhöfer, Jan
Master thesis
Thumbnail
View/Open
no.ntnu:inspera:146039120:130868025.pdf (63.05Mb)
URI
https://hdl.handle.net/11250/3108849
Date
2023
Metadata
Show full item record
Collections
  • Institutt for maskinteknikk og produksjon [4307]
Abstract
 
 
Wood is a sought-after resource used in various application areas, representing sustainability and natural aesthetics. However, the wood industry and its secondary wood products face challenges of low levels of automation and quality issues. Due to high material costs, improvements in utilization are necessary to remain competitive in a volatile environment. This goal cannot be achieved through traditional manual quality controls. Computer vision approaches for defect detection in the wood industry hold great potential. Automated non-destructive technologies (NDT) can significantly enhance operational efficiency and lay the foundation for zero-defect manufacturing (ZDM) principles. In the context of this work, the effective application of image classification and object detection computer vision technologies for wood defect detection is examined. Furthermore, this thesis explores whether offline data augmentation and transfer learning are effective methods for improving performance with limited data quantities. To further evaluate these methods, new data is collected using a self-installed camera system in a simulated production environment. The results demonstrate that modern YOLOv7 and YOLOv8 one-stage object detectors outperform classic image classification algorithms in terms of overall performance and usability. Through the application of offline data augmentation and transfer learning, performance can be partly enhanced when working with limited data sets.
 
Publisher
NTNU

Contact Us | Send Feedback

Privacy policy
DSpace software copyright © 2002-2019  DuraSpace

Service from  Unit
 

 

Browse

ArchiveCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsDocument TypesJournalsThis CollectionBy Issue DateAuthorsTitlesSubjectsDocument TypesJournals

My Account

Login

Statistics

View Usage Statistics

Contact Us | Send Feedback

Privacy policy
DSpace software copyright © 2002-2019  DuraSpace

Service from  Unit