A framework for leaf disease analysis and estimation using MAML with DeepLabV3

Abstract

In the dynamic field of agriculture, the prompt and accurate identification of plant diseases plays a crucial role in ensuring strong crop yields. To address this need, we present an innovative framework that combines Model Agnostic Meta-Learning (MAML) with DeepLabV3 to identify plant diseases precisely and assess their severity. Deep Plant Guard utilizes the outstanding segmentation capabilities of DeepLabV3 to pinpoint areas of diseased plants while simultaneously determining the specific type of disease and its severity level. The framework is enhanced by Bayesian task augmentation-MAML with multi-scale spatial attention, allowing for swift fine-tuning even with limited data. During training, a composite loss function harmonizes segmentation and classification efforts. Following meta-training excels in adapting to new tasks, providing detailed segmentation masks, and offering valuable insights into disease type and severity. Evaluation results, based on datasets such as Plant Village, Plant Doc, and a newly introduced plant disease dataset, showcase impressive results, including a 99.1% accuracy rate, 99.5% sensitivity, and 98.7% specificity. These results highlight the framework's effectiveness in addressing disease types and severity assessments.