Towards Discrimination of Plant Species by Machine Vision: Advanced Statistical Analysis of Chlorophyll Fluorescence Transients

Automatic discrimination of plant species is required for precision farming and for advanced environmental protection. For this task, reflected sunlight has already been tested whereas fluorescence emission has been only scarcely considered. Here, we investigated the discriminative potential of chlorophyll fluorescence imaging in a case study using three closely related plant species of the family Lamiaceae. We compared discriminative potential of eight classifiers and four feature selection methods to identify the fluorescence parameters that can yield the highest contrast between the species. Three plant species: Ocimum basilicum, Origanum majorana and Origanum vulgare were grown separately as well as in pots where all three species were mixed. First, eight statistical classifiers were applied and tested in simulated species discrimination. The performance of the Quadratic Discriminant Classifier was found to be the most efficient. This classifier was further applied in combination with four different methods of feature selection. The Sequential Forward Floating Selection was found as the most efficient method for selecting the best performing subset of fluorescence images. The ability of the combinatorial statistical techniques for discriminating the species was also compared to the resolving power of conventional fluorescence parameters and found to be more efficient.