December 5, 2025. Chris Bidinosti, University of Winnipeg
The meeting welcomed new members and featured presentations from various research groups focusing on agricultural data collection, imaging, and machine learning applications. The discussions covered data storage and sharing systems, including a data portal hosting large agricultural datasets and potential collaborations with various organizations. The group addressed data access policies, research security, and the importance of data sovereignty in agricultural research, with emphasis on protecting Canada's competitive advantage while facilitating innovation and collaboration.
Summary
New Members and Plant Imaging Research
The meeting began with ARWG hosting in Dan's absence and welcoming three new members: Jean Christophe Ruel from Inno OP tech, Ryan McCarty from Myconics, and Howard Sinclair Jones. Christopher Bidinosti from the University of Winnipeg presented on their research group, Terabyte, which focuses on agriculture and digital imaging. They discussed their data generation methods, including a 3-axis gantry system for collecting labeled images, and their work on developing algorithms and 3D models for plant imaging. Michael will speak about data curation and hosting, emphasizing the importance of proper data management in plant sciences.
Data Portal Development Overview
Michael presented an overview of the data storage and sharing system, highlighting the collaboration with the Digital Research Alliance of Canada and the creation of a data portal hosting large datasets from EagleEye and field data. He explained the functionality of the portal, including the ability to filter and download specific images, and discussed potential future developments such as integrating cloud-based model training and expanding the types of data available on the portal. Michael also mentioned a potential collaboration with the International Weed Recognition Consortium and the involvement of Saiwa in evolving the portal further.
Automated Solutions for Agricultural Challenges
Christopher Henry presented Terabyte's work in digital agriculture, focusing on connecting with researchers to develop automated solutions for agricultural problems. He discussed their efforts in creating label datasets and machine learning models, as well as their work on synthetic data generation. The team is currently developing a system to automatically detect and assess fusarium head blight on wheat, collaborating with Maria Antonia at the Morden nursery. They are also working on weed detection and classification, as well as herbicide resistance studies, using automated segmentation methods for drone data.
Plant Protection and Yield Prediction
Christopher Henry discussed a project involving RNAi sequences in plants to control flea beetles, where the team needs to assess trial efficacy and count plant material after beetle interaction. He also described a three-year data collection project with Emily's Innovation Farm aimed at predicting pea yield and biomass, involving root nodule detection methods. Finally, he mentioned working with graph and hypergraph neural networks for the Canadian Grain Commission's FHB occurrence data analysis.
Agricultural Data Analysis Innovations
Christopher Henry presented various agricultural data analysis and machine learning applications, including work with the Canadian Grain Commission to forecast fusarium head blight using hypergraph methods, and a photogrammetry system to assess wheat droopiness levels. He demonstrated a computer vision approach for assessing fungal leaf blights with 10% error compared to human assessment, and discussed remote sensing projects with Natural Resources Canada for enhancing digital elevation data resolution. The presentation concluded with a discussion of synthetic data generation using GANs and diffusion models for creating realistic agricultural imagery, with applications in weed detection and fusarium head blight analysis.
Synthetic Data for Agricultural Imaging
Christopher Henry discussed the use of synthetic data generation in agricultural imaging, addressing concerns about data variability and bias. He explained that diffusion models aim to capture data distributions and that the team is working on determining the minimum amount of real data needed for specific applications. Anne-Michelle, representing Agriculture and Agri-Food Canada, expressed interest in connecting Christopher's work with her team's projects involving satellite data and quantum computing for agricultural modeling. They discussed the potential for multi-scale analysis combining satellite, drone, and plant-level data, with opportunities for collaboration.
Agricultural Data Access and Collaboration
The meeting discussed data access and collaboration for agricultural research. Christopher Henry clarified that the data is released under Creative Commons license to encourage innovation, and Michael explained the straightforward process for obtaining an account to download the data. David Hobson suggested limiting database access to Canadian researchers and companies, emphasizing the importance of establishing formal collaborations to ensure mutual benefit and protect Canada's competitive advantage in agricultural data.
Data Access and Sharing Policies
The group discussed data access and sharing policies for their portal, with Michael explaining that accounts are required for data access and they conduct case-by-case reviews of data requests. Christopher Henry described their data management plan which initially restricted access to researchers and then industry, while Christopher Bidinosti noted they have turned down data requests from companies like Picture This. The discussion concluded with a detailed explanation from Christopher Henry about the varying amounts of data needed for different machine learning applications, ranging from hundreds to hundreds of thousands of images depending on the complexity of the problem.
Research Security and Data Sharing
Ron Van Holst from Compute Ontario joined the meeting to discuss research security and data sharing in Canada. Christopher Henry explained that their data collection is curated with the intention of sharing, and they have protocols in place for secure data storage and destruction. They also discussed the importance of data sovereignty and the need to address dual-use technology concerns, with Christopher noting that they have plans in place to address these issues.
