U.S. researchers say they’ve pinpointed individual spud plants infected with potato virus Y with 90 per cent accuracy, using hyperspectral cameras mounted on drones.
Donna Delparte, an assistant professor of geosciences at Idaho State University, and graduate student Mike Griffel have successfully tested a “computer-learning” algorithm they developed to tease out PVY from spectral imaging “background noise,” such as field variability and unrelated crop stress.
“Our premise was to look at all of these wavelengths of light the human eye can’t see and look for differences between healthy plants and plants infected with PVY,” Griffel said, adding their images had leaf-scale resolution.
Griffel said the project detected disease well before potato crops reached the row-closure stage, far earlier than people can spot symptoms of PVY by scouting fields.
To develop their algorithm, they compiled crop data in fields over three seasons, ending in 2016. The researchers first analyzed fields from the ground with a high-tech camera capable of recording 100 bands of the light spectrum.
After studying the images, they selected the 15 most useful bands for identifying PVY based on its unique light reflection. Delparte programmed more basic hyperspectral cameras mounted on drones to detect those bands while surveying the same potato fields from the air.
They developed the algorithm based on common spectral signatures among sick plants. Their software “learned” to ignore field variability based on comparisons of sick plant signatures with signatures reflected from adjacent healthy plants.
Griffel envisions the technology will eventually enable drones to text GPS coordinates of sick plants to field agronomists, or direct drones to spray and kill sick plants upon detection.
Source: Capital Press