Aircraft Wake Vortex Recognition Method Based on Improved Inception-VGG16 Hybrid Network.

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Computer Vision

Background:

• Accurate identification of aircraft wake vortices is crucial for aviation safety.
• Traditional methods struggle with complex background noise and multi-scale features.
• Deep learning offers potential for enhanced wake vortex detection.

Purpose of the Study:

• To propose and validate a hybrid deep learning network, Inception-VGG16, for accurate aircraft wake vortex identification.
• To enhance feature extraction capabilities for multi-scale wake vortex characteristics.
• To improve the robustness of wake vortex detection in complex environments.

Main Methods:

• Developed a hybrid Inception-VGG16 network architecture.
• Utilized a Feature0 module for preliminary feature extraction from Doppler radar velocity data.
• Employed improved InceptionB and InceptionC modules for multi-scale feature extraction.
• Integrated VGG16's hierarchical structure for deep feature extraction and classification.

Main Results:

• The Inception-VGG16 model achieved a classification accuracy of 98.8% on 3530 wind field samples.
• Significantly outperformed traditional methods (SVM, KNN, RF) and the single VGG16 network.
• Demonstrated superior ability in processing multi-scale wake features and handling complex backgrounds.

Conclusions:

• The proposed Inception-VGG16 hybrid model provides a highly accurate and robust solution for aircraft wake vortex identification.
• This deep learning approach enhances aviation safety monitoring systems.
• The model effectively overcomes limitations of single networks in complex feature extraction.