Classification and detection of powdery mildew damage levels in cucurbits plants through spectral analysis

Abstract

In this work, a methodology to detect three damage levels by powdery mildew on cucurbit plants through spectral signatures is proposed. Leaves in fungal germination, leaves with the first symptoms and diseased leaves are considered. A database of spectral signatures with the wavelengths grouped into regions of interest is used. The feature extraction and data reduction is determined with an exploratory analysis and principal component analysis. The wavelength bands with the most descriptive data are between the ranges of 405-679 nm for the visible band and 760-988 nm for the near infrared band. Then, the detection of damage levels is obtained, with a classification accuracy of 93.2 % and a Cohen's kappa value of 0.81. This spectral analysis provides a model that allows features between damage stages for powdery mildew in the cucurbits plants.