Recognition in the early stage of powdery mildew damage for cucurbits plants using spectral signatures

Abstract

One of the most significant diseases affecting cucurbit plants is powdery mildew, which causes substantial yield losses in both greenhouses and field crops, especially during the winter and summer periods. Therefore, early diagnosis and detection are essential for effective pathogen control. An advanced, non-invasive method was developed for remotely sensing this fungal disease and assessing damage levels using spectral reflectance. The primary objective of this study is to detect the onset of the disease before the first visible symptoms appear on the leaves through the use of vegetation indices. To achieve this, statistical analyses and multiple comparison tests were employed for feature selection, in combination with machine learning algorithms, such as a Support Vector Machine. The results demonstrated high reliability in distinguishing between healthy and infected cucurbit leaves with powdery mildew. By calculating vegetation indices (VIs), seven optimal features were identified, enabling the recognition of three damage levels with 98% accuracy and a Cohen's $\kappa$ coefficient of up to 0.96. Spectral reflectance successfully differentiated powdery mildew damage levels in cucurbit plants, suggesting that this method could be recommended for crops with similar characteristics.