Automatic detection of 'yellow rust' in wheat using reflectance measurements and neural networks

Excessive use of pesticides for plant disease treatment increases costs and raises the danger of toxic residue levels on agricultural products. As pesticides are among the highest components in the production costs of field crops and have been identified as a major contributor to ground water contamination, their use must be minimised. This can be achieved by more precise targeting of pesticides to those places in the field where they are needed. Therefore, a simple and cost-effective optical device is needed for remote disease detection, based on canopy reflectance in several wavebands. In this study, the difference in spectral reflectance between healthy and diseased wheat plants was investigated at an early stage in the development of the "yellow rust" disease. In-field spectral images were taken with a spectrograph mounted at spray boom level. A normalisation method based on reflectance and light intensity adjustments was developed. An innovative technique for visualisation of data properties and interrelations between variables is presented, based on Self-Organizing Maps. Disease detection algorithms were developed, based on neural networks. Through the use of neural networks and more specifically multilayer perceptrons, classification performance increased from 95% to more than 99% using a total of 5137 leaf spectra for evaluation. These results encourage prospects for the development of a cost-effective optical device for recognising diseases at an early stage. (C) 2004 Elsevier B.V. All rights reserved.