Depending on environmental factors fungal diseases of crops are often distributed heterogeneously in fields. Precision
agriculture in plant protection implies a targeted fungicide application adjusted these field heterogeneities. Therefore an
understanding of the spatial and temporal occurrence of pathogens is elementary. As shown in previous studies, remote
sensing techniques can be used to detect and observe spectral anomalies in the field. In 2008, a sugar beet field site was
observed at different growth stages of the crop using different remote sensing techniques. The experimental field site
consisted of two treatments. One plot was sprayed with a fungicide to avoid fungal infections. In order to obtain sugar
beet plants infected with foliar diseases the other plot was not sprayed. Remote sensing data were acquired from the
high-resolution airborne hyperspectral imaging ROSIS in July 2008 at sugar beet growth stage 39 and from the HyMap
sensor systems in August 2008 at sugar beet growth stage 45, respectively. Additionally hyperspectral signatures of
diseased and non-diseased sugar beet plants were measured with a non-imaging hand held spectroradiometer at growth
stage 49 in September. Ground truth data, in particular disease severity were collected at 50 sampling points in the field.
Changes of reflection rates were related to disease severity increasing with time. Erysiphe betae causing powdery
mildew was the most frequent leaf pathogen. A classification of healthy and diseased sugar beets in the field was
possible by using hyperspectral vegetation indices calculated from canopy reflectance.
The aim of this research is the early detection of plant diseases based on the combination of vegetation indices. We have
seen that an individual index such as the most popular one, namely NDVI, does not discriminate adequately between
healthy and diseased plants, e.g. Cercospora beticola, Erysiphe betae, and Uromyces betae. However, by combining
vegetation indices, which are usually called features in classification, very reliable results can be achieved. We use
Support Vector Machines for classification. By this we receive a classification accuracy of almost 95% for Cercospora
beticola and Uromyces betae and still over 92% for Erysiphe betae. Depending on the different plant diseases we have
found that different vegetation indices are important, too. Consequently, the question how to find the best index for every
plant disease and the choice of the best subset arise. Both questions are not the same, because different indices contain
similar information which can already be seen from the formula of the calculation of the vegetation index. These
dependencies do not have to be linear. In order to identify optimal subsets of features for the different pathogens already
at an early stage of infestation, we have found that entropy and mutual information are adequate concepts. Accordingly
we use the minimum redundancy - maximum relevance (mRMR) criterion to evaluate the features. We have found that
we need different indices and feature subsets of different sizes for different diseases.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.