Value Generation with Computer Vision
Techniques in the Olive Oil Industry -
a Knowledge Graph Framework
Quality assessments of olive oils are labour-intensive and time-consuming tasks for olive
oil producers. In addition to their own quality targets, producers have to meet extensive
standardised quality criteria in order to be allowed to market high-quality olive oils. For
the necessary chemical and organoleptic tests, various expensive measuring instruments
and extensive procedures are necessary; producers usually only resort to some internal
measurements and use their own industry knowledge for their decisions. Recent developments
combine the measurement method of fluorescence spectroscopy with machine
learning to analyse olive oil test samples in a applicable and fast way. Despite first promising
prediction and classification results, producers are left with questions about how
to interpret and use the results for their own production. To promote the explainability
of AI systems and knowledge alignment with industry knowledge, knowledge graphs
provide contextual understanding and improved data integration.
Subsequently, this thesis investigated how image recognition techniques for fluorescence
images of olive samples can be applied to enrich a domain-specific knowledge graph that
ultimately supports quality assessment in the olive oil industry. For this purpose, using
a design science research approach, results of fluorescent spectroscopy with the help of
computer vision and machine learning results, as well as the domain characteristics in
the quality assessment of olive oils were investigated. As a result, a domain-specific
KG was created, which was enriched using object detection, colour analysis and image
classification results based on a YOLOv8 and linear regression model. Olive oil producers
receive explanations about the results of the fluorescence images through the reasoning
of the knowledge graph and can identify dependencies to their own quality testing,
production and market management. The proposed framework supports the traceability
and overview of quality attributes and evaluation criteria by combining computer vision
and knowledge graph technologies. The basic concept provides a practical starting point
for industries relying on fluorescence spectroscopy and that are confronted with AI to
foster knowledge management for quality assurance.
Schmid, Christian, 2023
Art der Arbeit Master Thesis
Auftraggebende
Betreuende Dozierende Laurenzi, Emanuele
Keywords
Views: 22 - Downloads: 0
Studiengang: Business Information Systems (Master)
Vertraulichkeit: öffentlich