Researchers at James Cook University are combining machine learning and satellite imagery to detect sugarcane diseases before they develop.
Under the leadership of Professor Mostafa Rahimi Azghadi, the team developed a software tool that effectively differentiates between healthy and infected sugarcane.
“RSD can reduce sugar yields by as much as 60% and spreads rapidly. However, due to its asymptomatic nature, it is not detectable to the naked eye until late in the growing season,” Professor Azghadi explained.
Traditionally, RSD is diagnosed by manually cutting and sampling sugarcane, followed by DNA analysis of the samples in a laboratory.
“This process is labor-intensive and costly, especially when scaled up, as each test costs about $10 to $15,” Professor Azghadi pointed out.
“Our method achieved an accuracy of 86% to 97% depending on the sugarcane variety, which is comparable to or better than existing crop disease detection tools.”
In this study, we leveraged vegetation indices derived from publicly available Sentinel-2 data and used various machine learning techniques to identify RSD across different sugarcane varieties.
True samples were collected from 76 sugarcane blocks in the Herbert region of Queensland. This dataset was collected by trained agronomists at Herbert Cane Productivity Services.
The findings show that machine learning algorithms can “successfully classify RSDs across multiple types using freely available multispectral satellite data.”
“The ultimate goal is to create an early warning system that assesses disease risk while monitoring the overall health of crops and facilitates better management of agricultural productivity.”
“It works similarly to a regular medical check-up at a general practitioner, but it is tailored to sugarcane and other crops.”
