Rice pest identification based on multi-scale double-branch GAN-ResNet

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Rice production is vital for global food security, necessitating efficient pest and disease management.
• Smart agriculture relies on timely detection and prevention of crop issues.
• Deep learning shows promise for image-based pest identification but struggles with small datasets and complex natural environments.

Purpose of the Study:

• To develop an advanced deep learning model for accurate rice pest and disease identification.
• To address challenges of small datasets, overfitting, and feature extraction in natural environments.
• To improve the robustness and generalization of crop pest identification systems.

Main Methods:

• A Multi-Scale Dual-branch model integrating generative adversarial networks and an improved ResNet architecture was proposed.
• The model incorporated ConvNeXt residual blocks and a dual-branch structure for multi-scale feature extraction.
• Data augmentation techniques expanded a 5,932-image dataset to 20,000 images for training.

Main Results:

• The proposed model achieved a recognition accuracy of 99.34%, outperforming classical networks like AlexNet, VGG, DenseNet, ResNet, and Transformer.
• Compared to the original ResNet model, the new model demonstrated a 2.66% improvement in recognition accuracy.
• The model exhibited strong generalization ability and robustness in complex natural environments.

Conclusions:

• The Multi-Scale Dual-branch model offers a superior solution for rice pest and disease identification, overcoming existing limitations.
• This approach significantly enhances recognition accuracy and addresses overfitting issues caused by small datasets.
• The developed model provides a robust tool for smart agriculture, contributing to improved crop management and food security.