A Spectrum Recognition Method for Birds and Drones Using Residual Attention Networks

The classification of birds and drones has become a hot topic in recent years. Drones have been rapidly abused due to safety issues, and birds frequently threaten aircraft take-off and landing. But birds and drones share lower radar cross-section (RCS), speed, and altitude, making it difficult for r...

Full description

Saved in:
Bibliographic Details
Published in2024 9th International Conference on Intelligent Computing and Signal Processing (ICSP) pp. 728 - 732
Main Authors Chen, Jinfan, Mo, Junxian, Zheng, Zixiang, Xie, Hanfeng, Zhang, Yue
Format Conference Proceeding
LanguageEnglish
Published IEEE 19.04.2024
Subjects
Birds
birds and drones
Doppler effect
Doppler radar
Doppler spectrum
Drones
Radar
Radar cross-sections
residual attention network
Residual neural networks
Safety
Signal processing
target classification
Vectors
Online AccessGet full text

Cover

More Information
Summary:The classification of birds and drones has become a hot topic in recent years. Drones have been rapidly abused due to safety issues, and birds frequently threaten aircraft take-off and landing. But birds and drones share lower radar cross-section (RCS), speed, and altitude, making it difficult for radar to distinguish. This article uses residual attention networks to classify radar multi frame spectral sequence data of drones and birds. Firstly, Fourier transform (FFT) is used to transform the radar target to obtain the Doppler spectrum vector of the target. A multi frame Doppler dataset of birds is constructed by cutting. Due to the scarcity of drone radar data, a sliding window method is used to construct a multi frame Doppler dataset of drones. Then, due to the attention module's ability to obtain global and local connections, which conform to the characteristics of radar targets in the Doppler spectrum, we combined it with residual neural networks to solve the classification problem of birds and drones using residual attention networks. The experiment shows that the recognition rate of this method on Doppler datasets can reach over 96%.
DOI:10.1109/ICSP62122.2024.10743946