Investment

GRDC Code: UOQ2306-005RSX

GRS (Chrisbin James) Virtual Agricultural Imaging and Sensing through Artificial Intelligence and Computer Vision

Phenotyping grain count (and grain count per panicle) in sorghum has been traditionally performed via threshing. Grain count is a trait that is highly correlated with grain yield and, additionally, grain count per panicle is also associated with traits crop growth per unit area and criterion for selection under high-temperature conditions. In the first year of this PhD project, a deep learning algorithm was successfully developed that uses a combination of three images and a rough 3D model of a sorghum panicle to estimate grain count on panicles collected from breeding trials. This year, the project is focused on developing methodologies to identify the panicle morphology based on canopy architecture measured in the field. UAV-based 3D reconstruction techniques were reviewed and tested, which use low-altitude flights for imaging canopies. Initial results show the quality and shape of individual panicles derived from UAV-based 3D reconstruction correlated well with the measured panicle shape in lab. Training deep learning models to identify panicle shapes requires large amounts of labelled 3D datasets. A 3D sorghum canopy model generation framework is also being developed, based on panicle shape data collected in the first year and publicly available 3D sorghum datasets to simulate UAV and LiDAR 3D data.

Project start date:

01/07/2023

Project end date:

31/12/2025

Crop type:

Not Crop Specific

Organisation

The University of Queensland

Region:

North, South, West

Project status

Active

GRDC News

GroundCover Supplement

Grains research recruits from a wide science...

1726149600000

The opportunity to acquire and analyse big data from new technology such as drones, satellites...

Feedback