CVPPP Meeting: Sept 10th 2015

GARNet took on a change of pace recently with attendance to a workshop on Computer Vision Problems in Plant Phenotyping (CVPPP), which was arranged as part of the British Machine Vision conference. The majority of attendees were computer scientists who had taken on the interesting and challenging problem of measuring different aspects of plant growth.

In the workshop introduction, the co-organiser Sotirios Tsaftaris highlighted that the ability to automate phenotyping is key to the rapid selection of new crop varieties, many of which will be critical for feeding the worlds growing population. This was highlighted in recent talk at the COPO workshop where Chris Rawlings from Rothamstead introduced a phenotyping field imager that takes 1000s of images of plants grown in the field. However to get anything out of this huge production of data, the phenotypes need to be rapidly documented and analysed. When one imagines that phenotypes can be altered by both genetic variation and climatic conditions (amongst others) then it is a significant challenge to take consistent measurements and makes plain why problems with computer vision software can be a major bottleneck.
BottleneckImage

The majority of the talks at the meeting presented software programs to measure leaves in an Arabidopsis rosette and I was surprised to observe the variety of tactics that were employed for this ostensibly simple task. The main tactic employed was to use segmentation analysis to remove the complexity of a RGB image before attempting to count the leaf number. The processing of these images was conducted by a variety of different computational tactics such as ‘finger-counting’, ‘Gaussian-process-shape modeling’ or ‘fuzzy c-means’. Although I didn’t understand the detail of these programs, the researchers had had some success, although this seemed limited to plants with ~10 rosette leaves. Arguably the most complete method can be found on the PhenoTiki software, although this website still requires addition of some detailed content.

Outside of the leaf counting work there was an excellent talk by Hanno Scharr from the Julich Plant Phenotyping Centre (which havephenoSeeder a wide range of available tools) who introduced their phenoSeeder machine that aims to allow ‘3D Reconstruction of Seeds by Volume Carving’. This machine is not high throughput and is confused by raised areas on the seed but Dr Scharr demonstrated proof of principle examples that showed the method is well-suited for 3D seed phenotyping.

 

 

Overall it seemed to me that the majority of the talks described methods that were still in their infancy even though they had taken significantly computational time to generate the models. However this research area is well served with tools. The excellent Plant-Image-Analysis.org describes 130+ programs that have been designed for or can be used for different aspects of plant phenotyping. Each researcher should certainly proceed with caution when using any of these programs to ensure that they are working well for their specific needs.

PIA

The UK phenotyping community is in a great position both with the ongoing UKPPN and the UK Plant Phenomics Centre in Aberystwyth, which is always looking for new collaborators. In fact there is currently a call for pilot projects with a closing date of September 25th.

Check out the GARNet Phenomics page for more details of available facilities!

Leave a Reply


 © 2024 - Weeding the Gems