Evenemanget har passerat

Examenspresentation av August Löfgren

Title: Development of machine learning algorithms for plant leaf segmentation and health status classification based on chlorophyll fluorescence images


Evenemanget har passerat


During the photosynthesis in leaves some energy is reemitted, called chlorophyll fluorescence. The signal of chlorophyll fluorescence can be analysed and by examining the dynamic response induced by a step change in light, plant health information can be extracted. The method has previously been successfully applied in detecting abiotic stress. However for powdery mildew, a biotic stress, further research was needed. This thesis will examine the possibility of early disease detection of powdery mildew on strawberry plants using this method.

A fluorescence camera with spatial resolution is used to analyse leaves of the plants during the step change in search for features indicating biotic stress caused by powdery mildew. Machine learning algorithms are developed for plant leaf segmentation and disease status classification. For this purpose a data set of fluorescence image series of strawberry plants is annotated with mask and class label.

Segmentation of leaves worked well despite the limited size of the data set and the machine learning method, Mask R-CNN, was suitable for this dataset. The analysis did not show the expected result of infected leaves having a step response that differed from the healthy reference leaves in strawberry plants. No features in the fluorescence information could be stated as consistent, meaning that powdery mildew did not affect photosynthesis as strong as other stresses. Consequently, no indication was found for possibilities of early disease detection of powdery mildew before or after visual symptoms occur.


August and Torsten