• Research

CropQuant for hybrid wheat

Image based machine learning technologies to enable the quantification of key traits during wheat flowering.

Project summary.

Led by: Zhou Group

Start date: January 2018

End date: March 2019

Duration: 15 months

Grants: Bayer Grant4Traits Research Funding, Bayer Foundation

Value: £55,780

Presently, measuring key traits for hybrid wheat is extremely labour intensive and highly subjective, as the evaluation is mainly carried out visually by the human eye. Hence, for hybrid wheat programmes, breeders urgently request enabling technologies to facilitate a scalable and reliable evaluation of these traits in field trials in the context of hybrid seed production. In doing so, breeders can obtain a better understanding of flowering duration and the cross-pollination capability of wheat in order to answer questions such as the necessary number of plants (male vs female) in mixed planting and whether a high number of spikes/spikelets on non-male sterile lines carries an advantage for cross-pollination.

Together with the hybrid wheat seed production breeding team at Bayer (which has been transferred to BASF), we are utilising crop phenotyping technologies developed by the Zhou laboratory to carry out time-lapse crop photography in the greenhouse and in the field. After that, state-of-the-art computing technologies (including crop imaging, remote sensing, computer vision, machine-learning and data analysis) are being developed based on the existing CropQuant analysis pipeline to enable trait analyses that are important for hybrid wheat seed production breeding programme. Besides hybrid seed production, the project could also be used to monitor leaf diseases such as stripe rust, as well as key growth stages critical for wheat hybrid breeding such as duration of flowering.


The project is divided into three stages:

1) Improve the CropQuant technology to collect high-quality wheat growth images in the greenhouse and in the field. Considerable detail such as anthers on wheat ears and spikelet will be retained even in extremely bright (caused by sunshine) or dark (caused by shadow or clouds) lighting conditions. Our preliminary experiments suggest that the above approach could also help us to resolve issues such as colour distortion (caused by sunlight during the summer) and low picture clarity (caused by strong wind or heavy rainfall) in the field.

2) Apply advanced image analysis and machine learning algorithms to detect anther extrusion over time. The feature detection will generate floral-related trait measurements that enable us to study changes in floral development like anther extrusion and spike architecture over time. This stage will be carried out indoor first and then the algorithm will be tested and improved by infield data to obtain a more accurate and consistent analysis of anther extrusion over time.

3) Apply a deep learning based analysis solution to quantify spikes per unit area and spikelet/spike numbers in field trials. After that, project partners will verify and improve the algorithm with reference data assessed manually and/or by using devices like Opto-Agri or Marvin, pictures generated using UAVs, and visual scoring of anther extrusion during flowering.


Dr Michael Schmolke

BASF project leader, Head of European Wheat Improvement

Impact statement.

We work closely with Dr M. Schmolke at the European Wheat Breeding Centre, Crop Science Division, Bayer AG (now transferred to BASF). We aim to provide a hardware and software toolkit for Breeding & Trait Develop in the hybrid wheat seed production breeding programme at BASF.

People working on the project.

Related reading.