Towards precision agriculture: metaheuristic model compression for enhanced pest recognition

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Crop diseases and pests cause significant agricultural yield losses.
• Traditional pest identification methods are labor-intensive and lack robustness.
• Existing deep learning models for pest recognition are computationally demanding and large.

Purpose of the Study:

• To develop a novel and efficient deep learning framework for accurate crop pest and disease identification.
• To improve the computational efficiency and feasibility of deep learning models in agriculture.

Main Methods:

• Utilized InceptionV3 as a backbone for feature extraction.
• Incorporated a channel attention (CA) mechanism for feature refinement.
• Employed a metaheuristic optimization algorithm to reduce model complexity and computational overhead.

Main Results:

• The proposed model achieved high performance on the CropDP-181 dataset.
• Demonstrated superior classification accuracy and computational efficiency compared to state-of-the-art methods.
• Achieved a precision of 0.932, recall of 0.891, F1-score of 0.911, accuracy of 88.50%, and MCC of 0.816.

Conclusions:

• The developed deep learning framework is effective for accurate crop pest and disease identification.
• The model shows significant practical potential for real-time agricultural monitoring systems.
• The approach offers a balance between high performance and computational efficiency.