Image classification and identification for rice leaf diseases based on improved WOACW_SimpleNet

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Manual hyperparameter tuning for Convolutional Neural Networks (CNNs) is inefficient and costly.
• Accurate identification of rice leaf diseases is crucial for crop yield and food security.

Purpose of the Study:

• To develop an automated method for optimizing CNN hyperparameters for rice leaf disease identification.
• To improve the performance and efficiency of CNN models in agricultural applications.

Main Methods:

• Proposed a novel optimization algorithm: nonlinear convergence factor and weight cooperative self-mapping chaos optimization algorithm (WOACW).
• Incorporated opposition-based learning for population initialization and a polynomial mutation operator for enhanced optimization.
• Applied WOACW to optimize hyperparameters (learning rate, batch size, kernel size/number) for a CNN model (WOACW_SimpleNet).

Main Results:

• WOACW demonstrated superior performance compared to five other improved whale optimization algorithms on benchmark functions.
• The WOACW_SimpleNet achieved an average recognition accuracy of 99.35% for identifying five types of rice leaf diseases.
• An F1-score of 99.36% was recorded, indicating high precision and recall in disease classification.

Conclusions:

• The proposed WOACW algorithm effectively optimizes CNN hyperparameters, overcoming limitations of manual tuning.
• WOACW-optimized models show exceptional accuracy and efficiency in identifying critical rice leaf diseases.
• This approach holds significant potential for advancing precision agriculture and crop disease management.