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.

Abstract. In potato seed production in Japan, plants infected by viruses must be rogued from the field to provide commercial certified seed tubers. Seed producers detect infected plants over the field by inspecting visual symptoms. This roguing process is time-consuming, and it is generally difficult to identify them owing to mild symptoms. Therefore, an alternative way to effectively detect infected plants is desired. The objective of this research is to develop a detection method of virus-infected plants in the potato seed production field using an image classification technique. RGB images are taken at an altitude from 5 m to 10 m from the ground using an unmanned aerial vehicle (UAV). A total of 130 images of infected plants and 1300 images of healthy plants are collected. The number of images of infected plants is increased to 1300 by rotating the original images. Of the 2600 images in total, the dataset of 1800 images, which equally includes infected and healthy plants, is used as a training set, while the remaining 800 images are used as a validation set. The convolutional neural network (CNN) is applied for classification between infected and healthy plants. After parameter tuning of the CNN, the accuracy of the classification with the training data is 96%, while that with the validation data is 84%.