Rice-Disease Severity Level Estimation Using Deep Convolutional Neural Network

Severity of rice disease is an important indicator for farmers to plan appropriate treatments for protecting diseases which may give damages to rice paddy fields. This paper proposes an estimation method of rice-disease severity level using deep convolutional neural network. Since rice-disease severity level is the ratio between rice disease area and whole area in a rice leaf, candidate boundaries of rice disease are detected, and those boundaries are classified into a level out of rice-leaf disease levels which are early, middle, and final stages. All classified boundaries of levels are calculated with total area, and finally ratio with the whole leaf area is obtained. To evaluate performance of the proposed method, experiments with 2,500 images and 5,000 disease boundaries including five disease types have been performed, and results reveal 96.40%, 96.40%, and 96.56% accuracy for early, middle, and final stages, respectively.