Boundary-Based Rice-Leaf-Disease Classification and Severity Level Estimation for Automatic Insecticide Injection

Highlights A rice-leaf-disease detection and classification algorithm for multiple rice-leaf-diseases in a complicated rice leaf image is proposed this article. To increase accuracy, an coarse-to-fine determination proposed. Since features of types such as color, shape, so on are similar difficult to classify even with the human eye, tolerances among those small. The considers enlarging using two-step coarse-to-fine. Severity level disease also estimated our method. Abstract. Farmers may decide select appropriate insecticide rice-leaf treatment paddy field based class severity level. estimate field, several parts included captured image, sometimes there exists more than one boundary part leaf. This article proposes method estimation diseases image. first finds candidate boundaries identifies its feature area ratio. enlarge tolerance concept, categorized into two major groups coarse level, then both classified classes fine evaluate performance method, experiments were performed 8,303 images three including brown spot, blast, hispa healthy leaf, achieved 99.27% which outperformed deep learning approach by 0.43%. Keywords: Coarse fine, Multiple diseases, Rice-leaf recognition,