Early detection of grapevine downy mildew using thermal imaging

The objective of this research was to detect downy mildew in grapevine leaves at early stages of development using thermal imaging technology based on the assumption that plant disease causes significant modifications in leaf temperature. Infected and healthy leaves of grapevine grown in a controlled greenhouse experiment were classified using thermal images that were acquired 1, 2, 4 and 7 days from infection. Leaves were detected using the active contour algorithm for edge detection. The resulting leaf mask was used to calculate 14 features. Stepwise regression analysis revealed four significant features that were consequently used in prediction models. Five models were developed to classify between infected and healthy leaves. The best results were obtained by the support vector machine model using cross-validation on all data with a classification accuracy of 69.2% and an F1 score of 74.9%. Identifying downy mildew 4 and 7 days after infection resulted in an accuracy of 66.4% and 83.1% respectively.