Machine Learning for Remote Sensing of Xylella fastidiosa

Javier Galván Fraile (Advisors: Ramasco, Jose J.; Matias, Manuel A.)
Master Thesis (2020)

Xylella fastidiosa (Xf) is a plant pest able to infect over 500 plant species worldwide. This
pathogen has already caused considerable economic and environmental damage to olive
groves in Apulia (Italy) in recent years, and has since spread throughout Mediterranean
coastal zones. However, there are no effective control strategies against it and the Eu-
ropean Commission currently establishes hard eradication measures in the some of the
most affected regions. Particularly, all susceptible plants that are within a radius of 100
meters around an infected specimen must be uprooted, resulting in a great economic loss.
Consequently, diverse techniques and methods have been developed to detect the presence
of Xf in crops and monitor its spatio-temporal spreading dynamics in a
large scale in order to prevent its expansion and impact. Traditional infield survey methods
are accurate but costly for regional studies and monitoring. Instead, remote sensing along
with machine learning algorithms constitute a quick and cost-effective methodology for
determining the presence of the disease. Hence, in this project we present a novel tech-
nique for automatic detection of Xf from satellite imagery. Particularly, we
employ WorldView-2 satellite imagery with their 8-band multispectral data and a selection
of vegetation indices for the purpose of training selected machine learning algorithms
(SVM, artificial neural networks, recurrent neural networks, etc.) to determine whether an
almond tree has the disease or not. The pilot testing has been carried out in Son Cotoner
d’Avall farm (Puigpunyent, Mallorca), where a sample of 749 almond trees have been
subjected to q-PCR tests for Xf during 2018, wherefrom we are provided
with a WorldView-2 satellite image dated 22 June 2011. The applied multidisciplinary
approach is promising, as the trained algorithms show accuracies above 65% despite of
the time lag between the Xf tests and the satellite image. Therefore, this
work shows that large-scale satellite Xf monitoring is feasible and opens the possibility of
significant and promising progress based on this idea.


Esta web utiliza cookies para la recolección de datos con un propósito estadístico. Si continúas navegando, significa que aceptas la instalación de las cookies.


Más información De acuerdo