Enhanced convolutional block attention module with Learnable Gated Fusion (LGF-CBAM) for cocoa pod disease...

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Accurate cocoa pod disease detection is crucial for sustainable agriculture and minimizing yield loss.
• Deep learning models show potential but face challenges in feature extraction and generalization across datasets.

Purpose of the Study:

• To introduce a novel attention-based deep learning framework, LGF-CBAM with ResNetV2-101, for enhanced cocoa pod disease classification.
• To improve discriminative feature learning and model robustness in identifying cocoa diseases.

Main Methods:

• Developed a Learnable Gated Fusion Convolutional Block Attention Module (LGF-CBAM) that adaptively balances spatial and channel attention.
• Integrated LGF-CBAM with a ResNetV2-101 backbone for improved feature learning.
• Employed a softmax function for normalizing trainable gating parameters.

Main Results:

• Achieved 98.95% accuracy, 99.11% F1, and 99.11% PPV on the Cocoa_Pod_Disease_Gh dataset.
• Demonstrated cross-dataset robustness with high accuracies (94.00%-98.53%) on diverse datasets.
• Outperformed existing systems in accuracy, stability, and generalization for cocoa disease detection.

Conclusions:

• LGF-CBAM represents a state-of-the-art approach for accurate and robust cocoa pod disease identification.
• The framework offers a scalable solution for early and reliable disease detection, benefiting precision agriculture.
• This research contributes a novel attention mechanism for deep learning in plant pathology.