Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Detection of foliar disease in the field by the fusion of measurements made by optical sensors

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan

Citation:  Paper number  023087,  2002 ASAE Annual Meeting . (doi: 10.13031/2013.10454) @2002
Authors:   Cédric Bravo, Dimitrios Moshou, Roberto Oberti, Jon West, Alastair McCartney, Luigi Bodria, Herman Ramon
Keywords:   Spectrograph, imaging fluorescence, hyperspectral imaging, disease detection, sensor fusion, neural networks, Yellow Rust, Puccinia striiformis

The objective of this research was to detect and recognize plant stress caused by disease in field conditions by combining hyperspectral reflection information between 450 and 900nm and fluorescence imaging. The aim is to develop a tractor mounted cost-effective optical device for site-specific pesticide application in order to reduce and optimize pesticide use. The work reported here used yellow rust (Puccinia striiformis) disease of winter wheat as a model system.

In the field hyperspectral reflection images of healthy and infected plants were taken with an imaging spectrograph mounted at spray boom height. Leaf recognition and spectral normalization procedures to account for differences in canopy architecture and spectral illumination were used. A model, based on quadratic discrimination, was built, using a selected group of wavebands to differentiate diseased from healthy plants. The model could discriminate diseased from healthy crop with an error of about 10% using measurements from only three wavebands.

Multispectral fluorescence images were taken on the same plants using UV-blue excitation. Through comparison of the 550 and 690 nm fluorescence images, it was possible to clearly detect disease presence. The fraction of pixels in one image, recognized as diseased, was set as final fluorescence disease variable called the lesion index LI).

The lesion index was added to the pool of normalized selected reflection wavebands. This pool of observations was used in a quadratic discrimination model. This model was further refined using a neural network approach. The combined model improved disease discrimination compared to either the spectral model or fluorescent model and had a classification error of between 1 and 2 %.

The results suggest that there is potential for developing detection systems based on multisensor measurements that can be used to in precision disease control systems for use in arable crops.

(Download PDF)    (Export to EndNotes)