Attention-Based Deep Convolutional Neural Networks for Plant Disease Classification

Abstract

Abstract Plant diseases pose a significant threat to global food security and agricultural productivity. In this work, we propose a novel deep convolutional neural network (CNN) model enhanced with Squeeze-and-Excitation (SE) blocks and Attention Gates (AGs) for multi-class plant disease classification across five crops: apple, maize, grape, potato, and tomato. Leveraging a large image dataset and a comprehensive training regime, the proposed model achieves high performance across all metrics, including 99% accuracy, 0.99 F1-score, and strong specificity. Evaluation includes feature visualization and Grad-CAM interpretability. The model's robustness and interpretability make it a compelling solution for practical agricultural applications.

Plant phenotyping relevance

植物画像から病害状態を分類するCNN手法の開発と性能評価が中心であり、植物病害フェノタイピングに該当する。

Code and data availability

The paper's plant disease classification experiments are built directly on the public PlantVillage Kaggle image dataset (21 classes across five crops), which is the paper-specific image input for its phenotyping measurements. No author analysis code, trained model checkpoints, or other paper-specific assets are stated.