Early detection of bacterial canker in tomato plants using spectroscopy for smart agriculture applications

Abstract

Clavibacter michiganensis subsp. michiganensis (Cmm) causes bacterial canker in tomatoes, causing severe yield loss. It would be of practical research interest within smart agriculture to develop an effective and quick method to distinguish pre-symptomatic infected tomato plants from healthy ones to take protective measures in time. In this study, artificially inoculated tomato plants with Cmm were grown in a temperature-controlled chamber. Using the Relief method, 25 wavelengths in the visible spectrum (cyan and red regions) showed the highest statistical differences, between healthy and asymptomatic infected tomato plants, two days before the first appearance of the foliar symptoms, in each plant. In addition, inoculated tomato plantlets showed differences in contrast to healthy ones, in the near-infrared spectrum, thirteen days after the inoculation with Cmm. The spectral data were used for the creation of early detection models of healthy and inoculated pre-symptomatic plants, in a specific number of days before the appearance of the first symptoms, in each plant, and in a specific number of days-post inoculation, using two ML algorithms (SVMs and kNN). The algorithms proved effective and robust in the discrimination of the two classes of the two instances mentioned. Furthermore, three patterns of data-preprocessing followed before the training of the algorithms, i.e. the case of multidimensionality, the application of PCA, and the use of Relief method. Finally, six models were created for datasets that contain spectral data of asymptomatic inoculated with Cmm and healthy tomato transplants, all of which showed very high overall accuracy, ranging from 92 to 100%.