Automated estimation of tiller number in wheat by ribbon detection

Abstract: The advent of high-throughput phenotyping installations signals a need for plant biology to use pat- tern analysis and recognition techniques, especially when analysis is done via digital images. Such installations also provide an opportunity to computer vision. We describe one such application at the UK National Plant Phenomics Cen- tre, in which historically measurements have been made in a labour-intensive manual manner. We develop an estimator of tiller number in growing wheat which, when exploit- ing per-day averaging, temporal interpolation and dynamic programming, delivers measurements of finer-grain and no less accuracy than manually, and provides observations on plant treatments hitherto difficult or impossible to obtain. The approach developed lends itself to reuse for any similar imaging setup, and plants with tillering characteristics sim- ilar to wheat. We consider the work a useful exemplar for co-operation between biologists and computer scientists in such installations.

Click Here