Detecting Variety Contamination in Sugarcane Plots

Cross contamination between different sugarcane varieties is shown to increase the risk of a disease outbreak. Current methods of detection are unable to quickly and accurately identify areas of risk over large areas. By utilising drone images, spectral reflectance data can be used to discriminate the differences in light between each variety. Multispectral data will be captured in both visible and near infrared light bands. Multiple vegetation indices will also be utilised to increase the predictive power of both models. This research project aims to build a machine learning and statistical model that is capable of detecting different varieties of sugarcane over large areas. The classification model will be trained on the eight different varieties of sugarcane and will be capable of detecting multiple varieties within a single plot. The statistical model will be created as an anomaly detection model that will find the probability of contamination within a plot based on its surroundings.

Project Team and Collaborators:

JCU Team: Dr. Carla Ewels, Connor Whiteside and Ethan Water
Industry Partner: Rob Milla and Rina Patane of Burdekin Productivity Services