A machine learning approach to drought stress level classification of tobacco plants

We show that a model for drought stress level classification of tobacco leaves can be learned from measurement data. The data was acquired using a sheet-of-light measurement system developed at the Fraunhofer Institute for Integrated Circuits IIS. Spatial attributes like length, width or bending were extracted by fitting a parameterized leaf model to the measurement data. The attributes were transformed to simple attribute vectors describing relevant aspects of plant growth and stress evidence. The resulting attribute vectors were used to train decision trees, neural networks and linear regression classifiers. To provide a broad range of data, plants were assessed in a planned measurement campaign. Stress was induced by cutting off the water supply to simulate drought. Evidence for drought stress could be recognized from the data. Classification of whole plants yielded better results than classification of single leaves.