Published on Tuesday, March 1, 2022
The research department at Beck’s continues to grow and mature, and our breeders’ appetite for more phenotyping data points to be collected in the yield trial fields each growing season grows with it. Hybrid phenotypes are the visual observations of the plant resulting from the interaction of its genotype with the environment. Given the many different locations each testing hybrid is planted in during each testing season, the importance of understanding this interaction helps breeders make better product advancement decisions.
Traditionally, research testing associates are tasked to walk the fields every growing season, taking notes on each hybrid during specific growth stages. This process is incredibly time-consuming and takes a lot of people, training, logistics, planning. As we expand, it has become more difficult to cover the ground we need in a timely manner.
For a few years now, our research team has been flying unmanned aerial vehicles (UAVs) during the growing season to help standardize and increase the number of observation data points for traits like stand count and plant height. However, some important hybrid traits, like ear height, reside below the late-season canopy that UAVs can’t see. So, we must revert to more traditional methods of data collection, which means walking the fields with large measuring sticks to measure ear heights for each hybrid. In larger fields, there can be thousands of measurements which can take several days of work to complete.
Taking ear height measurements is a hot, dusty job, so, in 2018 we asked, “Is there a better way?” A computer science professor from IUPUI approached Beck’s, wanting to find a partner to contribute agricultural datasets for a class project on machine learning. Through the partnership, we presented the question to the students about whether plant ear height measurements could be collected using videos instead of the more traditional way. At the end of the semester, the students presented their machine learning model solutions for measuring ear heights using GoPro videos. One student’s work stood out, and Beck’s purchased it, laying the groundwork for the internal technology development project we call the “Camera on a Combine Project.”
For a second year, during the fall harvest of 2021, the research team tested collecting video footage from the head of their combines as they harvested research plots. This is an internal, crossfunctional project that incorporates the field expertise of the research team and the computer data science expertise of our technology team. The two teams have joined forces to ground truth and train a machine learning model to detect the ears on the stalks of corn using an object detection model. Then, it will measure the height of the ears using stereo vision properties in the combine video footage with impressive accuracy. These days, the combine camera hardware is a little more sophisticated than a GoPro camera. The machine learning model is a little more complex as we train it in new environments and lighting conditions, but the goal to replace the manual measurement work with a standardized, efficient high throughput way to collect hybrid ear height data remains the same.
After the second season of testing the combine mounted camera system in the field, the team is planning for a larger scale of systems in the field in 2022. The goal is to tackle any problems or make any improvements to the system before the 2023 launch of a fully developed automated solution for taking ear height measurements in the field.
Hopefully, one day, walking fields with a 12-foot measuring stick will be a thing of the past.
Author: Brandi Payne
Categories: CropTalk, 2022