University of Twente Student Theses


The improvement of objective milkfoam quality analysis through image processing and computer vision

Hendriks, Koen (2020) The improvement of objective milkfoam quality analysis through image processing and computer vision.

[img] PDF
Abstract:PCV Group is a product development agency with a specialization in dispensing and dosing systems of coffee machines. Such systems are mainly used in coffee and milk foam treatment. PCV’s systems are analysed intensively to ensure their quality. Currently, their milk foam analysing method is operator dependant which is not desirable in this field of work. To improve upon this, we researched the possibilities to improve the objectiveness of milk foam quality analysis through image processing and computer vision. We investigated the current analysing method in collaboration with an operator of PCV Group to determine the operator dependant actions and find design opportunities for a new method. The purpose of this new method is to correctly classify the fineness and distribution of milk foam samples with high reliability and reproducibility. We performed a literature research to acquire a better understanding of computer vision-based techniques that could be implemented in this analysing method. Common techniques recommended by the literature are mainly based on high detail imaging, contrast differences, and measuring bubble sizes through bubble segmentation. To explore which techniques could potentially contribute to this purpose we tested different combinations of imaging techniques, image optimisation techniques and image analysing techniques in an iterative designing process. We found that the segmentation techniques recommended by the literature did not meet our requirements. We investigated other segmentation techniques which led to colour thresholding and machine learning-based segmentation being the most promising segmentation techniques in this research. Colour thresholding and machine learning-based segmentation are used to exactly measure bubble sizes of milk foam samples, but classification in fineness and distribution is not yet integrated with these methods. The most promising classification technique is machine learning-based classification and, in this research, this is used to correctly classify 91% of the fineness scores and 82% of the distribution scores. Some improvements, like increasing training data and creating consistent imaging techniques, are needed before this method can be integrated into the milk foam analysis. However, the findings of this research can be used as a foundation for the design of a new milk foam analysing method.
Item Type:Essay (Bachelor)
PCV Group, Enschede, The Netherlands
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:50 technical science in general, 54 computer science, 58 process technology
Programme:Creative Technology BSc (50447)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page