Tip-burn stress detection of lettuce canopy grown in Plant Factories

A compelling effort has been made in recent years to face several kinds of plant stresses using a variety of sensors and deep learning methods. Yet most of the datasets are based on single leaves or on single plants, exhibiting explicit diseases. In this work we present a new method for stress detection which can deal with a dense canopy of plants, grown in Plant Factories under artificial lights. Our approach combining both classification and segmentation with self supervised masks, and WGAN based data augmentation, has the significant advantage of using normal rgb low cost cameras, simple data aquisition for training and it can both localize and detect the tip-burn stress on the plant canopy with very good accuracy as shown in the results. We have tested our results also on datasets available on tensorflow.org.