||Multispectral Analysis of Seeds
||Jensen, Anders Boeck
||Carstensen, Jens Michael (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||This thesis presents work done with multispectral image analysis of seeds with
the goal of finding fungus infections on the seeds. The goal is to determine if
a seed is infected and if so with which fungi it infected, and to what degree it
is infected. Besides the work with the seeds, it is also tested if it is possible to
remove camera refl
ections from a multispectral image.
The main approach used is spectral unmixing which decomposes the image
into a number of endmembers, each corresponding to clean seed or the fungi.
5 different fungi were chosen for this study. With the decomposition specific
fungus infection on the seed may be determined. The image data are extended
with morphological closing and Laplace filtering with the aim of getting better
results. Statistical classification is also used to identify the fungi in order to
compare spectral unmixing with a simple method.
The result with removing camera re
flection shows that spectral unmixing is not
suited for this task. The camera re
flection can be identified in the images, but
When applying spectral unmixing on the seed data, it is possible to find out if a
seed is infected or not. Using only the image data, however, it is not possible to
identify which fungus is infecting the seed. When also using an extension based
on morphological closing, it is possible to identify 3 of the 5 fungi studied. This
makes it possible to determine which fungus a whole seed is infected with if it
only has one infection, but the result was not good enough to determine the
fungi at a pixel level. Classification was able to identify all fungi, but once
more, not at a pixel level. With spectral unmixing it is possible to determine the degree of infection, while this is not possible with classification.
The results of this thesis show that it is possible to identify the fungi with both
spectral unmixing and classification, and they also show that spectral unmixing
is not superior to classification. More work is needed for both techniques to
work in practical applications.
||Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Creation date: 2009-03-11
Update date: 2009-11-04