Details
| Originalsprache | Englisch |
|---|---|
| Qualifikation | Doctor rerum horticulturarum |
| Gradverleihende Hochschule | |
| Betreut von |
|
| Datum der Verleihung des Grades | 18 Aug. 2023 |
| Erscheinungsort | Hannover |
| Publikationsstatus | Veröffentlicht - 2023 |
Abstract
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
Hannover, 2023. 117 S.
Publikation: Qualifikations-/Studienabschlussarbeit › Dissertation
}
TY - BOOK
T1 - Towards automated phenotyping in plant tissue culture
AU - Bethge, Hans Lukas
PY - 2023
Y1 - 2023
N2 - Plant in vitro culture techniques comprise important fundamental methods of modern plant research, propagation and breeding. Innovative scientific approaches to further develop the cultivation process, therefore, have the potential of far-reaching impact on many different areas. In particular, automation can increase efficiency of in vitro propagation, a domain currently con-strained by intensive manual labor. Automated phenotyping of plant in vitro culture bears the potential to extend the evaluation of in vitro plants from manual destructive endpoint measurements to continuous and objective digital quantification of plant traits. Consequently, this can lead to a better understanding of crucial developmental processes and will help to clarify the emergence of physiological disorders of plant in vitro cultures. The aim of this dissertation was to investigate and exemplify the potential of optical sensing methods and machine learning in plant in vitro culture from an interdisciplinary point of view. A novel robotic phenotyping system for automated, non-destructive, multi-dimensional in situ detection of plant traits based on low-cost sensor technology was con-ceptualized, developed and tested. Various sensor technologies, including an RGB camera, a laser distance sensor, a micro spectrometer, and a thermal camera, were applied partly for the first time under these challenging conditions and evaluated with respect to the resulting data quality and feasibility. In addition to the development of new dynamic, semi-automated data processing pipelines, the automatic acquisition of multisensory data across an entire subculture passage of plant in vitro cultures was demonstrated. This allowed novel time series images of different developmental processes of plant in vitro cultures and the emergence of physiological disorders to be captured in situ for the first time. The digital determination of relevant parameters such as projected plant area, average canopy height, and maximum plant height, was demonstrated, which can be used as critical descriptors of plant growth performance in vitro. In addition, a novel method of non-destructive quantification of media volume by depth data was developed which may allow monitoring of water uptake by plants and evaporation from the culture medium. The phenotyping system was used to investigate the etiology of the physiological growth anomaly hyperhydricity. Therefore, digital monitoring of the morphology and along with spectro-scopic studies of reflectance behavior over time were conducted. The new optical characteristics identified by classical spectral analysis, such as reduced reflectance and major absorption peaks of hyperhydricity in the SWIR region could be validated to be the main discriminating features by a trained support vector machine with a balanced accuracy of 84% on test set, demonstrating the feasibility of a spectral detection of hyperhydricity. In addition, an RGB image dataset was used for automated detection of hyperhydricity using deep neural networks. The high-performance metrics with precision of 83.8% and recall of 95.7% on test images underscore the presence of for detection sufficient number of discriminating features within the spatial RGB data, thus a second approach is proposed for automatic detection of hyperhydricity based on RGB images. The resulting multimodal sensor data sets of the robotic phenotyping system were tested as a supporting tool of an e-learning module in higher education to increase the digital skills in the field of sensing, data processing and data analysis, and evaluated by means of a student survey. This proof-of-concept study revealed an overall high level of acceptance and advocacy by students with 70% good to very good rating. However, with increased complexity of the learning task, stu-dents experienced excessive demands and rated the respective session lower. In summary, this study is expected to pave the way for increased use of automated sensor-based phenotyping in conjunction with machine learning in plant research and commercial mi-cropropagation in the future.
AB - Plant in vitro culture techniques comprise important fundamental methods of modern plant research, propagation and breeding. Innovative scientific approaches to further develop the cultivation process, therefore, have the potential of far-reaching impact on many different areas. In particular, automation can increase efficiency of in vitro propagation, a domain currently con-strained by intensive manual labor. Automated phenotyping of plant in vitro culture bears the potential to extend the evaluation of in vitro plants from manual destructive endpoint measurements to continuous and objective digital quantification of plant traits. Consequently, this can lead to a better understanding of crucial developmental processes and will help to clarify the emergence of physiological disorders of plant in vitro cultures. The aim of this dissertation was to investigate and exemplify the potential of optical sensing methods and machine learning in plant in vitro culture from an interdisciplinary point of view. A novel robotic phenotyping system for automated, non-destructive, multi-dimensional in situ detection of plant traits based on low-cost sensor technology was con-ceptualized, developed and tested. Various sensor technologies, including an RGB camera, a laser distance sensor, a micro spectrometer, and a thermal camera, were applied partly for the first time under these challenging conditions and evaluated with respect to the resulting data quality and feasibility. In addition to the development of new dynamic, semi-automated data processing pipelines, the automatic acquisition of multisensory data across an entire subculture passage of plant in vitro cultures was demonstrated. This allowed novel time series images of different developmental processes of plant in vitro cultures and the emergence of physiological disorders to be captured in situ for the first time. The digital determination of relevant parameters such as projected plant area, average canopy height, and maximum plant height, was demonstrated, which can be used as critical descriptors of plant growth performance in vitro. In addition, a novel method of non-destructive quantification of media volume by depth data was developed which may allow monitoring of water uptake by plants and evaporation from the culture medium. The phenotyping system was used to investigate the etiology of the physiological growth anomaly hyperhydricity. Therefore, digital monitoring of the morphology and along with spectro-scopic studies of reflectance behavior over time were conducted. The new optical characteristics identified by classical spectral analysis, such as reduced reflectance and major absorption peaks of hyperhydricity in the SWIR region could be validated to be the main discriminating features by a trained support vector machine with a balanced accuracy of 84% on test set, demonstrating the feasibility of a spectral detection of hyperhydricity. In addition, an RGB image dataset was used for automated detection of hyperhydricity using deep neural networks. The high-performance metrics with precision of 83.8% and recall of 95.7% on test images underscore the presence of for detection sufficient number of discriminating features within the spatial RGB data, thus a second approach is proposed for automatic detection of hyperhydricity based on RGB images. The resulting multimodal sensor data sets of the robotic phenotyping system were tested as a supporting tool of an e-learning module in higher education to increase the digital skills in the field of sensing, data processing and data analysis, and evaluated by means of a student survey. This proof-of-concept study revealed an overall high level of acceptance and advocacy by students with 70% good to very good rating. However, with increased complexity of the learning task, stu-dents experienced excessive demands and rated the respective session lower. In summary, this study is expected to pave the way for increased use of automated sensor-based phenotyping in conjunction with machine learning in plant research and commercial mi-cropropagation in the future.
U2 - 10.15488/14601
DO - 10.15488/14601
M3 - Doctoral thesis
CY - Hannover
ER -