ShuffleNetV2 SSM MLCA: a lightweight recognition network for wheat fungal diseases

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Wheat fungal diseases pose a significant threat to global food security, causing substantial yield losses.
• Manual disease diagnosis is labor-intensive and subjective, while current AI models lack accuracy in real-world settings.

Purpose of the Study:

• To develop a lightweight and accurate deep learning model for classifying wheat fungal diseases.
• To improve upon existing models' performance in complex field environments.

Main Methods:

• A novel convolutional neural network, ShuffleNetV2_SSM_MLCA, was constructed using SS-Conv-SSM modules and Mixed Local Channel Attention (MLCA).
• SS-Conv-SSM modules enhance feature extraction for similar-looking diseases, while MLCA focuses on discriminative features and reduces background noise.
• The model was trained and evaluated using standard configurations and five-fold cross-validation.

Main Results:

• The ShuffleNetV2_SSM_MLCA model achieved a classification accuracy of 91.35%, outperforming the baseline ShuffleNetV2 by 1.16%.
• Ablation studies confirmed that SS-Conv-SSM modules improved accuracy by 0.89%, and MLCA contributed an additional 0.27% increase.
• The model demonstrated a strong balance between computational efficiency and high classification performance.

Conclusions:

• The ShuffleNetV2_SSM_MLCA offers an efficient and precise solution for wheat fungal disease recognition.
• This technology supports real-time monitoring in intelligent agriculture, contributing to enhanced grain production and food security.