A STUDY IN CORN KERNEL DENSITY, MEASUREMENT, AND COUNTING FOR R&D w/SYNGENTA

OBJECTIVE

Farmwave kernel counting capabilities were conducted during a test run on 300+ ears of corn provided during a precision planting spacing trial. All images were processed through Farmwave’s kernel counting algorithm and examined by Farmwave development team.

APPROACH

Of the 369 ears, great variety was available from ears that had solid kernel formation and consistent size and density (252 ears), as compared to irregular ears that exemplified the field pressures of close rom spacing. these featured malformations, aborted kernels, stunted growth, and great variations of kernel shape and sizes (117 ears).

RESULTS

On “perfect” ears, Farmwave’s kernel count capabilities consistently provided results within +/- five percent (5%) accuracy around 98% consistency. For “perfect” ears, Farmwave sought to count whole kernels. Kernels on cylindrical edges are not counted independently as machine learning was taking the cylinder image and flattening it for processing.

On “irregular” ears, the Farmwave algorithm needed to “learn” how to recognize aborted kernels based on several factors: coloration, spatial, lighting, and shape. For all Farmwave automation and identification, further development of edge definition allowed kernels to be extracted from the image more accurately. Results were within +/- five percent (5%) accuracy with approximately 97% consistency.

KEY LEARNINGS

Continued progression on developing kernel counting for research environments would require the effort similar to that required for a different “application” as opposed to merely making small changes to a current model developed for use in sunlight and extreme lighting conditions focused on outdoor usage.