Title: Public datasets and machine learning algorithms for applications in digital agriculture 

Date: Wednesday 19, 2021
Time: 10:45 am to 11:25 am 

The development of new farming practices and technologies that take advantage of large amounts of data collected from a myriad of different sensors offers one of our best bets for a greener, more equitable future, with a stable supply of nutritious food for everyone. Whether one speaks of precision agriculture, digital agriculture, or Agriculture 4.0 (in reference to the anticipated fourth agricultural revolution), this movement is expected to bring critical innovations based on automated methods and algorithms that are dependent on large amounts of data. Similarly, a wide range of sensor and imaging technologies will be employed: standard RGB cameras, multi- and hyperspectral cameras, 3D laser scanning, and photogrammetry. As a result, digital ag will be driven by a full complement of data, across all pertinent scales and parameters, collected and acted upon by robots, autonomous vehicles (AV), drones, unmanned aerial vehicles, and satellites.  

This talk begins with our efforts to automatically generate millions of labeled RGB images for use in developing deep learning algorithms for digital agriculture applications. The talk will highlight a new portal being developed by the Enterprise Machine Intelligence and Learning Initiative (EMILI) and hosted on Compute Canada’s Cloud and Object Storage systems. The goal of this portal is to provide Canadian researchers and industry with access to our data and machine learning models to spur innovation in the agriculture sector. The talk will also highlight our approach to data generation and management. Current and future work in generating datasets based on multi- and hyperspectral cameras, 3D multispectral scanners, as well as associated deep learning models, will be discussed. The talk will end with our data collection plans for summer 2021 as well as our work with satellite data. 

Dr. Christopher Henry has held NSERC Discovery funding throughout his career. The focus of Dr. Henry’s work is the theory and application of machine learning, such as his work in developing machine learning data sets for digital agricultural applications. This has led to machine learning collaborations with Sightline Innovation, Northstar Robotics, and DecisionWorks. 

Dr. Henry has also pioneered approaches to classify pixels obtained from satellite images using deep neural networks developed for semantic segmentation for the creation of land-use/land-classification maps. The success of this work has led to industry research contracts and grants with GeoManitoba and Manitoba Hydro, as well as collaborations with researchers at the Norwegian University of Science and Technology. This work has generated a start-up company, DeepGeo, to commercialize this work. DeepGeo was created through the Creative Destructive Laboratory in the Rotman School of Management at the University of Toronto and secured its first contract with Manitoba Agricultural Services Corp on a project to perform crop yield prediction from satellite data. 

Dr. Henry has also been collaborating for many years on GPU-based computing initiatives, including collaborations with Manitoba Hydro, Ubisoft, and Manitoba Hydro International.  He has worked to establish a provincial consortium of researchers in high-performance computing that ran several roundtable meetings with industry as well as two conferences; he has co-founded the Applied Parallel Computing and Collaborative Research Laboratory; co-led the process to establish the University of Winnipeg as an NVIDIA GPU Education Centre.