Related Experiment Video
Updated: Sep 17, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
529
Swin Attention Augmented Residual Network: a fine-grained pest image recognition method.
Xiang Wang1, Zhiyong Xiao1, Zhaohong Deng1
1School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, China.
Frontiers in Plant Science
|July 4, 2025
Summary
Accurate pest identification is crucial for crop safety. A new Swin Transformer-based method, Swin-AARNet, improves pest recognition by enhancing feature extraction and integrating multi-scale information, showing superior performance on large datasets.
Area of Science:
- Agricultural Science
- Computer Vision
- Machine Learning
Background:
- Pest infestations cause significant crop losses and economic damage.
- Existing pest recognition methods struggle with fine-grained visual differences and background interference.
- Accurate pest identification is essential for crop safety and agricultural management.
Purpose of the Study:
- To propose an improved pest identification method addressing limitations of current approaches.
- To enhance the accuracy and efficiency of pest recognition in agricultural settings.
- To develop a robust model capable of distinguishing visually similar pests amidst complex backgrounds.
Main Methods:
- Developed Swin-AARNet (Attention Augmented Residual Network), a Swin Transformer-based architecture.
- Enhanced local feature extraction through a feature complementation mechanism.
- Integrated multi-scale information to mitigate fine-grained feature ambiguity.
Main Results:
- Achieved 78.77% accuracy on the IP102 pest dataset.
- Demonstrated high accuracy on citrus datasets: 82.17% on CPB and 99.48% on Li.
- Swin-AARNet showed superior performance compared to state-of-the-art models in pest image classification.
Conclusions:
- Swin-AARNet effectively distinguishes pests with similar appearances and is robust to complex backgrounds.
- The method shows strong potential for real-world agricultural applications, including crop monitoring and early warning systems.
- This approach significantly advances automated pest identification for improved agricultural safety.
