Paft-wpest: wolfberry pests fine-grained classification method based on generative self-supervised learning

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Fine-grained pest recognition is crucial for agricultural production safety.
• Challenges include subtle differences, variations, background noise, and limited data.

Purpose of the Study:

• To develop a generative self-supervised learning model for improved fine-grained pest recognition.
• To address limitations of existing pest recognition methods.

Main Methods:

• Proposed PAFT-WPest model using partial-convolution spatial attention.
• Incorporated channel semantic selection and frequency-domain modeling.
• Developed two wolfberry pest datasets and used continual pre-training.

Main Results:

• Achieved high accuracies on multiple public pest datasets (e.g., 98.70% on WPIT9K).
• Demonstrated strong performance on self-built wolfberry pest datasets (e.g., 97.82% on WP45).
• PAFT-WPest effectively improves recognition under complex backgrounds.

Conclusions:

• The PAFT-WPest model offers a feasible approach for agricultural pest monitoring and classification.
• Enhanced pest recognition capabilities contribute to precise pest control.
• The model's adaptability is improved through continual pre-training.