Stimator: A Method in Agriculture CPS Framework to Estimate Severity of Plant Diseases using Graph Neural Network

Plant diseases are a major factor affecting agriculture yield by accounting for approximately 20% of the total produce and pose a constant threat to global food security. Accordingly, many solutions have been proposed to detect the presence of disease and estimate severity using AI and ML. These techniques estimate the severity as the proportion of area that was affected by the disease. Instead, this article estimates the severity of disease as the harmonic mean of the percentage of area affected by the disease and the percentage of area the disease has spread to. It represents the data of diseased locations in farmland as a graph and performs estimation with the help of GNN. These insights can help a disease management A-CPS to automate its downstream tasks like resource allocation and targeted spraying to control the disease efficiently. The proposed method has been validated by training a GNN on images of apple leaves in which diseased parts have been identified and annotated. When presented with a new image, the trained model produced results resembling human visual estimation.