Aircraft Image Recognition Network Based on Hybrid Attention Mechanism.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Deep Learning

Background:

• Traditional aircraft recognition methods struggle with low precision due to diverse aircraft types, model similarities, and texture interference.
• Advancements in deep learning offer potential for improved image recognition tasks.
• Fine-grained visual categorization remains a challenge in computer vision.

Purpose of the Study:

• To propose a hybrid attention network model (BA-CNN) for enhanced aircraft recognition.
• To improve the precision rate of aircraft image classification.
• To address limitations of traditional methods in distinguishing similar aircraft models.

Main Methods:

• A two-channel ResNet-34 architecture was employed for feature extraction, increasing network depth for enhanced fine-grained capability.
• A hybrid attention mechanism, incorporating channel and spatial attention modules, was introduced within the network.
• The attention mechanism focused on local channel and spatial characteristics, reducing redundancy and enhancing feature learning.

Main Results:

• The BA-CNN model achieved a recognition precision rate of 89.2% on the FGVC-aircraft dataset.
• The proposed method demonstrated a significant improvement in recognition precision compared to the original model.
• The BA-CNN model outperformed most existing mainstream aircraft recognition approaches.

Conclusions:

• The hybrid attention network (BA-CNN) effectively enhances aircraft recognition precision.
• Integrating channel and spatial attention modules improves fine-grained feature extraction.
• The BA-CNN model presents a promising solution for accurate aircraft classification in computer vision.