Evaluating the efficacy of fungicides for wheat scab control by combined image processing technologies

Abstract

Wheat scab, a major disease in wheat worldwide, is primarily controlled by fungicide application. Traditional methods used to evaluate fungicide efficacy are involved with manual counting of infected wheat heads in the field, which is time-consuming and laborious, and requires professional knowledge. This study developed a new method that could automatically and effectively assess the efficacy of fungicides in the field. Unet++ network and Fuzzy C-Means algorithm combined with R-G method (subtracting the green band from the red band in each image) were used to segment whole wheat ears and associated diseased areas, respectively. Proposed convolutional neural network (CNN) and connected domain method were then used to count all wheat ears and diseased wheat ears, respectively. The disease incidence of wheat ear groups was graded and the efficacy of five fungicides was evaluated. The results of these analyses show that the efficacy of all the five different fungicides was predicted accurately. The errors between predicted and measured levels of diseased wheat ears are less than 5%. The new method developed here can effectively assess the efficacy of fungicides for control of wheat scab under field conditions.