Machine learning-based classification of powdery mildew severity on melon leaves

Abstract

Precision agriculture faces challenges related to plant disease detection. Plant phenotyping assesses the appearance to select the best genotypes that resist to varying environmental conditions via plant variety testing. In this process, official plant variety tests are currently performed in vitro by visual inspection of samples placed in a culture media. In this communication, we demonstrate the potential of a computer vision approach to perform such tests in a much faster and reproducible way. We highlight the benefit of fusing contrasts coming from front and back light. To the best of our knowledge, this is illustrated for the first time on the classification of the severity of the presence of a fungi, powdery mildew, on melon leaves with 95% of accuracy.