Automatic Rice Plant Disease Evaluation Method Based on Anomaly Detection and Deep Learning

In rice breeding process, the main barrier is disease scoring and evaluation. The conventional evaluation process is slow because of manual evaluation. Rice breeder has to evaluate disease resistance in rice individually. This work introduced a rice blast disease classification on collected dataset using convolutional neural networks. Instead of classifying an image containing the entire rice plant, a proposed method performs a classification on multiple image sections selected from the defect region. This approach shows that a classification network can identify a disease from a bunch of small sections better than using a resized image of acquired as input image.