TriAttnNet based deep learning model for automated cotton pest detection and disease classification

Area of Science:

• Agricultural Science
• Computer Science
• Biotechnology

Background:

• Cotton crop health is vital for global agriculture.
• Pest detection and disease classification are challenging due to limited data and feature redundancy.
• Effective monitoring systems are crucial for sustainable agriculture and food security.

Purpose of the Study:

• To develop a robust deep learning framework for detecting cotton plant pests and classifying diseases.
• To address challenges of limited datasets, class imbalance, and feature redundancy in agricultural image analysis.
• To provide a computationally feasible and interpretable solution for precision agriculture.

Main Methods:

• Image preprocessing using Gaussian blur filtering and Contrast Limited Adaptive Histogram Equalization (CLAHE).
• Spa-GAN-based data augmentation for generating realistic synthetic samples.
• Attention-Guided Multi-Scale Residual U-Net (AGMS-U-Net) for Region of Interest (RoI) segmentation.
• TriAttnNet feature extractor with spatial, channel, and contextual attention.
• Hybrid Mongoose Ray Chaotic Optimization (HMRCO) for parameter tuning.
• Classification layer with focal loss.

Main Results:

• The proposed TriAttnNet achieved 98.66% accuracy, 98.71% recall, and 98.81% F1-score.
• Outperformed state-of-the-art models like EfficientNetB1-CBAM and BERT-ResNet-PSO.
• Demonstrated effectiveness in overcoming data limitations and improving feature representation.

Conclusions:

• The developed deep learning model offers a powerful and interpretable solution for cotton pest and disease management.
• The framework supports precision agriculture and sustainable cotton crop health monitoring.
• The system is computationally feasible for practical implementation in agricultural settings.