Brugger, Anna: Deep Phenotyping of disease resistance based on hyperspectral imaging and data mining methods in high throughput. - Bonn, 2022. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-68709
@phdthesis{handle:20.500.11811/10517,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-68709,
author = {{Anna Brugger}},
title = {Deep Phenotyping of disease resistance based on hyperspectral imaging and data mining methods in high throughput},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2022,
month = dec,

note = {Hyperspectral imaging for plant phenotyping has become an established research method using the visible to short wave infrared range (400-2500 nm). This allows for the differentiation, identification, and quantification of foliar diseases by linking spectral reflectance changes to different substances and processes within leaves. Although a wide range of secondary plant metabolites is involved in susceptible and resistant plant-pathogen interactions, their influence on the spectral reflectance has not yet been studied as they feature their absorption maxima in the ultraviolet range (UV; 100-380 nm).
This study aimed to implement a new hyperspectral imaging set-up in the UV-range to investigate different plant-pathogen interactions and characterise them by changes of secondary plant metabolites. For this purpose, measurement protocols, as well as evaluation routines, were established. Due to the harmful influence of UV-light on plant tissue, the measurement conditions were studied and set accordingly. Thus, it could be ensured that spectral changes were not induced by the UV-light source but by plant-pathogen interactions. The new measurement system was compared to an established non-imaging system by measuring several reference substances, where the imaging sensor produced comparable data with an additional spatial resolution. Following, plant-pathogen interactions of susceptible and resistant barley genotypes (Hordeum vulgare) inoculated with powdery mildew (Blumeria graminis f.sp. hordei) were investigated, and it was possible to distinguish different interactions by spectral changes. The in parallel extracted and analysed pigments and flavonoids were checked for correlation with the recorded spectral changes. In addition, hyperspectral imaging in the UV-range allowed for the differentiation of two foliar diseases of sugar beet (Cercospora leaf spot and sugar beet rust)by linking spectral changes to changes of secondary plant metabolites, which were quantified by destructive high-pressure liquid chromatography and extraction procedures. A subsequent analysis of the recorded spectral data with deep learning algorithms allowed for the differentiation between diseases and between symptom classes. However, the use of deep neural networks to accelerate the analysis of complex hyperspectral imaging data has to be examined critically to avoid the use of confounding factors to achieve high classification results. This was evident in the classification of an extensive high-throughput data set within this work, where elements of the image background were used instead of the plant samples for classification purposes.
The results of this thesis highlight that hyperspectral measurements in the UV-range can be used to study different plant-pathogen interactions and therefore allow for estimations about changes in the plant’s metabolism. Parallel extraction and analysis of secondary plant metabolites and pigments enabled a connection between spectral changes and metabolite changes, verifying the potential of hyperspectral measurements in the UV-range for plant phenotyping. The interaction of the sensor and UV-illumination remains challenging as significant peaks were recorded in all measurements, which cannot be considered for the interpretation of plant-pathogen interactions. While deep learning algorithms permit the detection and identification of plant diseases, they must be critically examined to exclude confounding factors. However, hyperspectral measurements in the UV-range can already be used to distinguish between plant-pathogen interactions and further investigate the influence of foliar diseases on plant physiology and biochemistry.},

url = {https://hdl.handle.net/20.500.11811/10517}
}

Die folgenden Nutzungsbestimmungen sind mit dieser Ressource verbunden:

InCopyright