A multi-scale feature fusion neural network for multi-class disease classification on the maize leaf images

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Maize leaf diseases significantly impact global crop yields.
• Accurate disease identification is crucial for effective crop management.
• Current AI methods struggle with maize leaf disease classification due to image variations and limited data.

Purpose of the Study:

• To develop an advanced artificial intelligence model for accurate multi-type maize leaf disease classification.
• To overcome limitations of existing AI methods in maize disease detection.

Main Methods:

• Proposed a novel residual-based multi-scale network (MResNet) incorporating two residual subnets of varying scales.
• Employed a hybrid feature weight optimization method for feature map fusion.
• Validated the model on a dedicated maize leaf disease dataset.

Main Results:

• MResNet achieved a high accuracy of 97.45% in classifying maize leaf diseases.
• The model outperformed existing state-of-the-art methods.
• Generalization performance was confirmed across multiple datasets.

Conclusions:

• MResNet offers a robust and accurate solution for maize leaf disease classification.
• The study enhances AI's role in precision agriculture and plant disease management.
• Thermodynamic diagram analysis improved model interpretability.