Detection of Leek Rust Disease under Field Conditions Using Hyperspectral Proximal Sensing and Machine Learning

Rust disease is an important problem for leek cultivation worldwide. It reduces market value and in extreme cases destroys the entire harvest. Farmers have to resort periodical full-field fungicide applications prevent spread of disease, once every 1 5 weeks, depending on cultivar weather conditions. This implies economic cost farmer environmental society. Hyperspectral sensors been extensively used address this issue research, but their application field has limited a relatively low number crops, excluding leek, due high investment costs complex data gathering analysis associated with these sensors. To fill gap, methodology was developed detecting rust using hyperspectral proximal sensing combined supervised machine learning. First, library constructed containing 43,416 spectra waveband range 400–1000 nm, measured under Then, extensive evaluation 11 common classifiers performed scikit-learn learning Python, variety wavelength selection techniques preprocessing strategies. The best performing model (linear) logistic regression that able correctly classify accuracy 98.14%, reflectance values at 556 661 first derivative 511 nm. unlabelled images, confirming accurately symptoms. can be concluded results work are step towards mapping future research needed overcome certain challenges before variable rate adopted against disease.