Robust Fine-Grained Pest Classification via Boundary-Aware Attention and Growth-Stage Supervision

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate pest identification is vital for effective pest management and sustainable agriculture.
• Fine-grained pest classification faces challenges like low inter-class separability, high intra-class variability, and environmental interference.

Purpose of the Study:

• To develop an advanced method for fine-grained agricultural pest classification.
• To enhance feature learning and reduce background noise in pest images.
• To improve classification accuracy under real-world, cluttered field conditions.

Main Methods:

• A boundary-aware channel-spatial attention network was developed.
• The attention module strengthens fine-grained structural and boundary cues.
• Auxiliary label supervision based on pest growth stages was incorporated to model developmental variations.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art baselines on the IP102 dataset.
• Consistent improvements in classification accuracy were observed.
• The method effectively enhanced inter-class separability and intra-class consistency.

Conclusions:

• The boundary-aware attention network with growth-stage supervision offers a robust solution for fine-grained agricultural pest classification.
• This approach has significant potential for real-world pest management applications.
• Integrating attention mechanisms and developmental stage information improves classification accuracy in challenging environments.