S18 - Session P3 - Detection of fire blight in apple trees using hyperspectral imaging

Authors: Belal Gaci *, Florent Abdelghafour, Silvia Mas-Garcia, Marine Louagrant, Florence Verpont, Aude Moronvalle, Ryad Bendoula, Jean-Michel Roger

Fire blight caused by Erwinia amylovora is a major phytopathology affecting numerous species of the rosaceous family such as Pyrus spp and Malus spp. Rapid detection of its symptoms in the orchard is a major challenge for arboriculturists. To control the disease and contain its spread, epidemiological surveillance procedures must be set up in the orchard. In this context, hyperspectral imaging coupled with chemometrics tools is an appropriate solution to detect fire blight symptoms. In this study, a SPECIM IQ hyperspectral camera was used in greenhouse conditions, to acquire images on grafted apple plants that were artificially inoculated. To process these images, a new method has been developed to predict diseased and healthy plants. This method exploits both the spatial and spectral properties of a hyperspectral image. The principle of this method consists, first, in extracting from the images, samples in the form of regions of interest (ROI) i.e. of a set of neighbouring pixels. Then, computing the spatial and spectral properties of the ROI. Finally, the two types of data are jointly modelled within the ROSA-PLS multi-blocks method. To apply this method, six images acquired 12 days after the inoculation were used, among which four images were used to establish the calibration function. From these images, both 80 infected and healthy ROI's of size 11*11 pixels were extracted. The calibration function was then tested on 30 ROI's extracted from independent test images. In this test, 100% of the samples are classified in their true corresponding class.