Rowbot receives commercialization support from the National Science Foundation
We are delighted to announce that Rowbot was recently selected for a Phase II SBIR award by the National Science Foundation (NSF). This two-year grant will support commercialization of a computer vision and machine learning based method (or AI) for guiding nitrogen (N) fertilizer application rate decisions for corn production.
Why does this matter? For corn farmers, N fertilizer is typically the most expensive input after seed purchase. And, importantly, N availability generally has the largest impact on yield after Mother Nature (weather). Yet, even today in 2024, it is commonplace for agronomists to estimate that farmers loose 20 or 30% or more of their applied N fertilizer. Those losses lower profitability while also creating water and air pollution challenges that undercut a grower’s stewardship.
Our novel approach begins with creating small areas of N stress using our proprietary robots. Some weeks later, sensors onboard our robots are used to evaluate leaf yellowness on lower leaves—a condition no farmer wants to see, but it is okay in this case because these conditions are created on tiny areas across a normal field. The images below show examples of characteristic yellowness (N stress) and how computer vision and machine learning are used to identify individual leaves. Yellowness scoring is carried out using another computer model (and validated by human scoring of many images). Maps of N stress can be created for test plots, which at this stage were narrow strips the entire length of a 10-ac field.
Finally, these observations of N stress are compared with predicted N stress from an N model to understand how well the N model matches actual field conditions. When a good match is made, then the model can be used to predict the final N application rate for the field. We believe that this approach will deliver higher profits, yields, and stewardship for growers across the Corn Belt.
Examples of N stress observed as characteristic yellowness on lower leaves, from no N stress (left) to high N stress (right).
Example of images processed using computer vision and machine learning to identify individual leaves that are subsequently scored for characteristic yellowness, a key indicator of N stress. "Segmented" leaves are shown in different colors (e.g., partial leaves in dark blue and fully-captured leaves in bright green).
Preliminary results from test field in Rosemount, MN. Nitrogen stress values were estimated using the computer vision and machine learning approach described above. There was good agreement with actual N stress imposed by varying application rates of N fertilizer (N-treatment codes ranged from no stress (0) to high stress (7).