A Graph-Related High-Order Neural Network Architecture via Feature Aggregation Enhancement for Identification...

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Plant diseases and pests pose significant threats to agricultural production, food security, and biodiversity.
• Accurate recognition of these threats remains a challenge for current information and intelligence technologies.
• Fine-grained visual classification is complex due to subtle inter-class similarities and intra-class variations.

Purpose of the Study:

• To develop an effective method for the fine-grained image recognition of plant pests and diseases.
• To address the limitations of traditional coarse-grained methods in distinguishing visually similar categories.
• To enhance the accuracy and efficiency of plant disease and pest identification systems.

Main Methods:

• Proposed a graph-related high-order network with feature aggregation enhancement (GHA-Net).
• Utilized an improved CSP-stage backbone network for multi-granularity feature extraction.
• Incorporated a multilevel attention mechanism for feature aggregation enhancement.
• Employed a graphic convolution module to analyze part-specific interrelationships in a high-order tensor space.

Main Results:

• GHA-Net demonstrated superior performance in accuracy and efficiency compared to existing models on public datasets.
• The network effectively captured robust contextual details for improved fine-grained identification.
• The approach proved suitable for complex scenes in plant disease and pest recognition applications.

Conclusions:

• The proposed GHA-Net offers a significant advancement in fine-grained visual classification for plant health monitoring.
• This method enhances the ability to accurately identify plant diseases and pests, supporting agricultural sustainability.
• GHA-Net provides a more effective solution for real-world applications requiring precise identification of agricultural threats.