A lightweight UAV recognition algorithm based on micro-Doppler features

The radar realizes unmanned aerial vehicle (UAV) recognition using the micro-Doppler effect caused by UAV rotors' rotation to extract micro-Doppler features of rotor echo signals. For example, principal component analysis (PCA) algorithm can extract features from the time-frequency spectrums obtained by the time-frequency analysis or its corresponding images. However, conventional frequency spectrums have a large amount of data, and PCA requires additional data dimension conversion when processing samples, prone to high covariance matrix dimensions and high computational complexity, which causes the time delay of feature extraction to increase exponentially. Therefore, in order to achieve lightweight and efficient individual recognition of small UAVs, this paper performs fast fourier transform (FFT) along the time dimension on spectrums, uses two-dimension principal component analysis (2DPCA) to reduce the data dimension to extract UAV micro-Doppler features, and send them to supervised learning classifiers to obtain the recognition results. The feature extraction algorithm takes a single sample as a calculation unit, which avoids high-dimensional data conversion, reduces computational complexity, and shortens the feature extraction time delay, with an average recognition rate of 98.44%.