Image-based phenotyping has become an indispensable part of efforts to quantify the effect of genotype and environment on important plant traits. Advances in automated imaging technology now enable researchers to collect large volumes of three dimensional data, captured under visible, fluorescent and/or near-infrared light. Processing such data sets in order to extract useful information regarding a plant’s structure, its state of development or specific functional properties requires an efficient computational platform.

The focus of this research group is to develop algorithms within an integrated software environment to enable the quantitative characterisation (phenotyping) of plant morphology, development and physiology, based on multimodal and multidimensional image data derived from both high-throughput greenhouse and field experiments, and microscopic, tomographic and three dimensional scanning studies.

The tasks include structure-preserving image enhancement, supervised and unsupervised image segmentation, registration, pattern recognition and classification, computational modelling and the description of shoots, leaves, roots, spikes, seeds, tissues and cells. Staff in the group collaborate closely with plant biologists, bioinformaticians and IT scientists.

• AVATARS - Advanced Virtuality and AugmenTed Reality AppRoaches in Seeds to Seeds (since 2019): development of algorithms for multimodal 2D/3D image registration, quantitative analysis and modeling of seed development.
   
• STARGATE - Sensors and daTA tRaininG towards high-performance Agri-food sysTEms (since 2021): collaboration and training in plant image analysis with EU partner organisations.
   
• DPPN - German Plant Phenotyping Network (2012-2019): development of algorithms and software solutions for automated shoot and root image analysis.
   
• Semi-automated plant image segmentation for efficient generation of ground truth data and precision phenotyping (since 2017).
   
• Fully automated plant image segmentation and phenotyping using pre-trained deep convolutional neural networks (since 2020).