Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : September 15, 2021 - September 17, 2021
Crop diseases are one of the factors behind the decrease in crop sales. Since the kind of crop diseases is large and there are diseases that damage other crops, early detection is required. Not only detecting disease strains needs much, but it is also difficult for new farmers to distinguish them; it is necessary to automate the early detection of diseased strains. This paper proposes a disease strain detection method using CAE (convolutional autoencoder) for onion strains. Since CAE learns only from strain images that are not disease, it is effective when there are few disease strain images. In the computer experiments, two types of image cropping methods are compared; the first type is to cut out manually the disease strain so that it appears in the center of the image. The second type is a method of automatically dividing images. As results of computer experiments, the effectiveness of the proposed method is confirmed; the discrimination rate is higher in the first cutting method than the second one.
