Machine learning based image analysis for phenotyping to speed up barley breeding

点击翻译按钮获取中文摘要

o Automated image-based plant phenotyping allows high throughput quantification of plant traits by analysing images captured by sensors and cameras in controlled environments or in the fields at any defined time interval. Visible light, fluorescent, near infrared, infrared and hyperspectral image can be captured from different viewing angles allowing the construction of 2D and 3D model of plants and their components, e.g. leaves, stems, flowers and spikes etc. Morphological and structure information about the plant, such as plant height and volumetric biomass, as well as physiological changes, such as temperature, water content, as well as the stress levels of leaves thus could be inferred, allowing non-destructive plant phenotyping in a high-throughput manner. Combined with experiments studying biotic and abiotic stresses, these images taken at multiple time points and conditions will illustrate the dynamic changes of the traits providing valuable information, such as the plant growth rate, stem elongation speed and trajectories of leave angle etc, thus providing key data for finding solutions for adaptation to climate change (e.g. drought, waterlogging, pathogen) and reduce the input for agriculture (e.g. fertilizers and pesticides).The proposed study will allow more accurate phenotyping with increased resolution and significantly reduced labour costs. This technology will greatly accelerate and enhance breeding of improved crops with beneficial architectural and physiological traits.