Interpretable deep multimodal-based tomato disease diagnosis and severity estimation.

Abstract

Plant diseases pose a significant threat to global food security, particularly in regions that rely heavily on crops that are vulnerable to disease, such as tomatoes. This research addresses the inefficiencies of traditional farming solutions by presenting a novel multimodal deep learning algorithm. The algorithm leverages EfficientNetB0 for image-based disease classification and utilizes Recurrent Neural Networks (RNN) to predict disease severity based on environmental data. By integrating visual and climatological inputs, our model addresses the limitations of unimodal systems, enhancing classification accuracy and interpretability. The model achieved a disease classification accuracy of 96.40% and a severity prediction accuracy of 99.20%. Additionally, the use of LIME and SHAP explainable AI techniques improves the understanding of disease severity classification outcomes. The contributions of this study align with precision agriculture practices and advance the resilience of local food systems, particularly in economies heavily dependent on tomato production. The proposed approach has the potential to mitigate the impacts of plant diseases and enhance food security by utilizing innovative technological solutions.

Plant phenotyping relevance

トマト病害の画像分類と病害重症度推定を行うマルチモーダル手法が研究の中心であり、植物状態の測定・推定に直接関わるため含める。

Code and data availability

The paper uses the public PlantVillage tomato leaf image dataset and a Kaggle weather dataset, but no authors' public URL for these inputs or for any analysis code/models is provided in the supplied blocks. The only URLs in the text (Kaggle weather dataset reference) are not among the allowed_urls, and no code deposit,

No evidence-backed public reproduction asset is currently recorded.