基于双通道GoogLeNet网络的旋翼无人机分类

针对利用雷达微多普勒效应的旋翼无人机识别问题，提出了一种基于双通道GoogLeNet网络的分类识别方法。首先对旋翼无人机的回波信号进行短时傅里叶变换（Short-Time Fourier Transform,STFT）从而获得信号时频谱，对时频谱沿时间轴进行傅里叶变换得到节奏速度图（Cadence-Velocity Diagram,CVD）。然后将时频图和CVD作为双通道GoogLeNet网络的输入进行特征提取用以获得回波信号的时频域和节奏速度域的特征，最后将所获得的特征输入到Softmax分类器中进而实现旋翼无人机的分类识别。基于实际雷达数据的实验结果表明，所提旋翼无人机分类方法准确率可达到98.9%。

Classification of rotor UAVs based on dual-channel GoogLeNet network

For the identification problem of rotary-wing unmanned aerial vehicle(UAV) using radar micro-Doppler effect,a classification and recognition method based on dual-channel GoogLeNet network is proposed.Firstly,Short Time Fourier Transform(STFT) is carried out to obtain the signal time spectrum of the micro-Doppler echo signal of the rotorcraft UAV.The Cadence-Velocity Diagram(CVD) is obtained by Fourier transform of the time spectrum along the time axis.Then,the time-frequency diagram and CVD are used as the input of the two-channel GoogLeNet network for feature extraction to obtain the characteristics of the time-frequency domain and the rhythm and velocity domain of the echo signal.Finally,the acquired features are input into Softmax classifier to realize the classification and recognition of the rotor UAV.Experimental results based on actual radar data show that the accuracy of the proposed classification method can reach 98.9%.