Imaging grain quality with AI

In order to define the quality of grain, Canada uses grain grades. Samples are taken to identify characteristics such as damage caused by insects, moisture, or temperature. Grain sampling is time-consuming and costly because it requires a person to take samples from grain flowing from an auger or combine at regular time intervals and deliver this to a lab for testing.

This project is part of our effort to advance the grain grading tools available to farmers and allow them to market their grains with confidence. The overall solution comprises several technology development initiatives that will be combined into a final product. Current research in machine vision tools show a high degree of effectiveness for the examining individual grain kernels.

Displaying the grain on a clear background or in an array has been repeatedly shown to differentiate types of grains, to identify specific wheat varieties, to count damaged kernels, and to identify quality factors like vitreousness. While effective in this environment, machine vision tools are challenged by the high throughput, in-field application that we require. In particular, machine vision struggles with identifying individual kernels when kernels are viewed in a mass.

Faculty Supervisor:

Tanya Lung;Susan Blum;Richard Dosselmann

Student:

Partner:

Ground Truth Agriculture

Discipline:

Computer science

Sector:

Agriculture; Information and cultural industries; Manufacturing

University:

Saskatchewan Polytechnic

Program: