AI studied as tool for sugar cane production

BATON ROUGE — Technological advances in artificial intelligence are making way for improved data collection for agricultural producers.
LSU AgCenter sugar cane breeder Collins Kimbeng has received a $288,690 award over three years from The John Deere Co. to study remote sensing and artificial intelligence for supporting sugar cane breeding and forecasting sugar yield.
“This project seeks to bring together expertise in artificial intelligence, precision agriculture and sugarcane breeding and also the benevolence of John Deere to solve two recalcitrant issues in the sugar industry,” he said.
The two goals of the study are to obtain accurate measurements of traits in the early stages of a sugar cane breeding program and the ability to forecast the best time to harvest a field to maximize sugar yield.
Kimbeng said the introduction of drones into sugar cane breeding will help his team work smarter rather than harder.
“Without this technology, we would have a number of people in the fields pulling samples to determine variety yield components,” he said.
Each year, the AgCenter Sugar Research Station examines more than 100,000 seedlings, of which only one or two will make it to the field as a sugarcane variety.
Kimbeng said a major problem is that the selection process is subjective.
“We put a lot of people in the field, and they all think they know what they are looking for,” he said. “But they could all have a personal bias.”
Cutting and weighing thousands of samples is time consuming. Kimbeng and his colleagues can now decrease the time and labor required with the use of technology.
“With this new remote sensing and artificial intelligence technology, we can take the weight of 100 samples,” he said. “We then input the information into the computer program and the drone can take the measurements we need with the help of onboard cameras.”
Drones will be used to acquire hyperspectral images from trials in the breeding program, including thousands of potential new varieties as well as fields planted with existing varieties.
“Data will be collected over several time series, and ground-truthing data will also be collected to accompany the drone data,” Kimbeng said.
“Statistical models will be developed to select the best predictive model containing the most appropriate wavelength to predict trait performance.”
If successful, this effort will lead to a more efficient selection process in the breeding program and a better way to predict sugar yield performance in the industry before harvest, he said.
Kimbeng will be joined in the project by AgCenter statistician Thanos Gentimis and precision agriculture expert Tri Setiyono.