About Flagship 4
Plant breeders select improved crop varieties from individual lines with desirable phenotypes. Techniques that increase the ability to select these desirable phenotypes accelerate the breeding process.
Image-based crop phenotyping allows the rapid and unbiased assessment of traits in the field that were previously difficult or impractical to measure, such as the rate of plant growth, the rate of plant ripening, or the intensity and duration of flowering. These image-based techniques can be used by plant breeders, geneticists and agronomists to optimize both plant breeding and crop management.
Flagship Project 4 combines image-based techniques (primarily drone images) and automated workflows to phenotype field-grown crops within plant breeding and precision agriculture programs. Field experiments can be imaged throughout the growing season using high-resolution visible light, multispectral and/or hyperspectral cameras, as well as novel imaging methods developed during the project. The ability to rapidly generate time-course images promises to increase the rate of genetic gain in crops and enhance the heritability of desired traits, and as a result, increase the efficiency of crop breeding programs. We are also extending the utility of phenotypes based on drone images to inform decisions on crop management, using satellite imagery to monitor crop development remotely, determine optimal input timing, and estimate crop yield. This also allows for the spatial optimization of crop inputs such as pesticides or nitrogen fertilizer to maximize productivity while avoiding environmental damage.
- Expanded capacity for larger-scale phenotyping
- Broader support for different imaging platforms and sensor types
- Improved understanding of image-based phenotypes that can lead to better automated analysis
- Improved capability for image-based phenotyping pipelines
- The plot explorer system allows researchers to investigate field trials in more detail than standard tools
The following projects are currently underway within Flagship 4:
Enhancing crop breeding by phenotype discovery and utilization through the development of high-throughput phenotyping methods to increase selection intensity.
Early-generation yield trials often lack replicates. We are using nested augmented experimental designs to test subsets of genetically related wheat and lentil populations for digital signatures that can be combined with genotypic data to predict the untested outcomes. We are also using structure from motion methods to estimate leaf area, plant height, and growth rate in structured lentil populations to identify genomic regions associated with growth rate. We are using image metrics for lentil herbicide injury to increase selection efficiency for resistance, and we are studying a structured population of canola to identify and quantify phenotypes contributing to nitrogen use efficiency.
We are in the first growing season for the nested augmented experimental designs and are collecting the first wheat and lentil image datasets. We have collected aerial imagery and ground-truth data from the Genome Canada AGILE project lentil populations (multiple years and locations) for genome-wide association studies to identify genomic regions controlling lentil crop volume growth trait phenotypes, and we will phenotype the canopy volume and growth rate based on imaging data. Ground-truth and genomic data have been collected from lentil plots at two locations with genetic entries for herbicide resistance, and field images will be used for the phenotyping of tolerance using a segmentation workflow for the odd-sized herbicide-tolerant lentil plots. We have assessed field-based phenotypic differences in nitrogen use efficiency among structured canola populations under high and low nitrogen conditions. The corresponding imaging data will be used to identify phenotypes that correlate with nitrogen use efficiency, which will be used by canola breeders.
Field Imaging for Precision Agriculture Applications
Extending phenotyping methods for the development of field imaging for precision agriculture applications.
We are measuring spectral shifts that occur as canola and lentil seedpods ripen, allowing the remote assessment of crop maturity without laborious manual inspection. We are also using image-based phenotypes as guides for in-season fertilizer application in canola, accounting for variations caused by unpredictable weather. We are using colour indices in aerial images for the field-scale estimation of canola yield using satellite images with a resolution of 3.5 m at ground level. We are also integrating diverse spatial information to remove variability from large-scale plant breeding plots during field experiments.
We are analyzing multiple canola and lentil varieties at multiple locations by ground-level sampling and hyperspectral imaging, and we have already identified hyperspectral phenotypes representing changes in seed color and moisture content that correlate with maturation. We are developing aerial phenotypes of canola that will guide the in-season application of nitrogen fertilizer based on a large trial at three locations testing the responses of two canola hybrids to the time and rate of fertilizer application. We have successfully developed methods to merge crop inventory classification with crop images, finding a clear relationship between maximum canola flowering and crop yield. We are integrating various forms of spatial information into large-scale field experiments, and developing models that optimize spatial covariance partitioning to maximize accuracy in field trials and precision agriculture.
Using novel sensors and processing methods for phenotype prospecting.
We are improving our ground-based phenotyping system for end-user deployment by making the hardware more robust and the interface more user-friendly. We are developing image analysis methods that can account for weather-induced variation without comparison to wet-dry reference surfaces. We are also developing new imaging-based methods to unlock the potential of ‘hard phenotypes’ that are related to climate change but cannot be accessed by manual testing or high-throughput phenotyping.
We have refined a second-generation ground-based platform (miniPAMM) to improve software reliability and data acquisition data extraction, and we have made it adaptable for specific end-user requirements. We are investigating phenotypes reflecting the response of plants to a changing environment, focusing on the impact of time-of-day effects on the sensitivity of thermal imaging in canola under different nitrogen treatments, as well as lentil domestication trials. We are developing strategies to measure ‘hard phenotypes’ based on leaf optical properties and deep learning algorithms. We have assembled a dataset of more than 300 leaves from a range of species broadly representing various leaf surface features and biochemistry. The spectral properties of lentil seed coats have been investigated in the context of breeding for seed coat type and resilience, and we are assessing optical coherence tomography as a means to measure seed coat thickness.
Image Pipeline Enhancements
Supporting rapid phenotypic data extraction and interpretation by enhancing the image pipeline.
The P2IRC image-processing pipeline is used by all P2IRC researchers and we are expanding it to improve access, functionality, and compatibility. We are developing software for interactive phenotype prospecting to support and record the exploration of P2IRC images, as well as interactive segmentation and a plot viewer application that allows users to inspect and explore both regular and irregular plots. Finally, we are expanding our plot explorer system to make it easier for researchers to evaluate and compare their field-trial data.
We have improved the image-processing pipeline to ensure faster uploading via a new interface, to allow background processing, to integrate new sensor and data types, and to accommodate a preview mode. We have developed reusable modules allowing the faster development of apps involving visual workspaces and visual interactions, and new interaction techniques for working with visual datasets, initially focusing on phenotypes associated with canola pods. The segmentation pipeline already incorporates blob-detection algorithms and we are now preparing irregular-plot datasets for further analysis. We have made substantial progress on the plot viewer application, including the development of features to support image retrieval and viewing, comparison within or between plots and growing seasons, assistance with data for publications, and integration with the image-processing pipeline.