Performance Assessment of Machine Learning Algorithms for Yellow Rust Wheat Disease Classification with UAV RGB Images

This paper proposes a workflow for the detection of yellow rust winter wheat disease from RGB images captured by Unmanned Aerial Vehicles, together with a sensitivity analysis against several influence quantities to Machine Learning (ML) classification algorithms working on these images. The study utilizes various image processing tools to extract meaningful features, and one-way ANOVA is employed to identify and prioritize critical features for disease diagnosis. Furthermore, ML algorithms are applied to these features to classify vegetation sickness, and the sensitivity analysis of the final ML model is evaluated against uncertainty sources such as blurring, light conditions, and noise affecting the acquired images. The model exhibits greater robustness to speckle noise when compared to other noise types while showing higher sensitivity towards contrast and light conditions in comparison.