Enhanced residual-attention deep neural network for disease classification in maize leaf images

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Accurate disease classification in maize is crucial for agricultural productivity and global food security.
• Deep learning, particularly convolutional neural networks (CNNs), shows significant promise for automated image-based classification tasks.

Purpose of the Study:

• To develop and evaluate MaizeNet, a novel CNN model for precise identification of diseases in maize leaves.
• To enhance model efficiency and performance using attention mechanisms and residual learning.

Main Methods:

• Development of MaizeNet, a CNN incorporating attention mechanisms and residual learning.
• Implementation of a five-fold cross-validation strategy for robust model generalization.
• Utilization of macro-average metrics to address class imbalance in the dataset.

Main Results:

• MaizeNet achieved high performance metrics: F1-score (0.9509), recall (0.9497), precision (0.9525), and accuracy (0.9595).
• The model demonstrated robustness and reliability in classifying maize leaf diseases.
• Attention and residual learning components contributed to improved gradient flow and feature focus.

Conclusions:

• MaizeNet offers a reliable and efficient automated solution for maize disease classification.
• The developed model supports timely agricultural interventions, enhancing crop yields and food sustainability.
• This research highlights the potential of advanced deep learning techniques in precision agriculture.