A deep learning-based framework for large-scale plant disease detection using big data analytics in precision agriculture

Abstract

Abstract Wheat and corn are essential crops for global food security, but wheat yellow rust and the corn northern leaf spot are significant threats. Proper assessment of the severity of the disease is the key to effective control and minimizing crop loss. Traditional methods don’t work effectively, and current deep learning models have problems like focusing too little on severity assessment, only being able to be used for a single crop or disease, and relying on small datasets, all of which make them less reliable in the real world. This paper addresses these issues. It introduces WY-CN-NASNetLarge, a deep-learning model based on the NASNetLarge architecture. The model is trained using transfer learning, fine-tuning, and several datasets, such as Yellow-Rust-19, Corn Disease and Severity (CD&S), and PlantVillage. These help the model work well in a variety of disease conditions. Data augmentation, the AdamW optimizer, dropout training, and mixed precision training enhance performance and prevent overfitting. The model has 97.33% accuracy for classifying disease severity. It is higher than ResNet152v2, InceptionResNetV2, and DenseNet201. This approach is effective and quick for identifying multiple diseases and rating their severity. It can also help manage diseases in agriculture and prevent crop loss.

Plant phenotyping relevance

植物病害の重症度を画像等から推定する深層学習手法の開発・比較検証が中心であり、植物の病害状態を直接評価するフェノタイピング研究に該当する。

Code and data availability

The paper trains its plant disease severity model on three public image datasets (Yellow-Rust-19, CD&S, PlantVillage) with explicit Kaggle/paperswithcode availability links. Two Kaggle-hosted datasets match allowed URLs exactly; the CD&S link in the text does not exactly match an allowed URL entry, and the authors' own