无人机被动音频探测和识别技术研究
Research on audio detection and recognition of UAV

为解决近场空域低、慢、小旋翼人机的安全威胁，提出基于音频信号分析的人机探测识别方法。该方法采用改进流程和参数的梅尔频率倒谱系数（Mel-Frequency Cepstral Coeffi-cients，MFCC）和其一阶差分作为人机音频的特征参数，结合提出的多距离分段采集法，通过训练高斯混合模型（Gaussian Mixture Model，GMM），建立多特征的人机音频“指纹库”，最后用特征匹配算法实现人机的探测和识别。实验结果表明，所提出的方法在典型郊区环境中可实现150 m距离内人机的探测和识别，识别率达到84.4%。
In order to solve the security threat from low, slow and small rotor UAV (unmanned aerial vehicle) in the near field airspace, a method of analyzing the UAV audio signal is proposed. In this method, the Mel-Frequency Cepstral Coefficients (MFCC) and its first-order difference with improved solving flowchart and optimized parameters are used as the characteristic parameters of the UAV audio signal. The multi-distance segmentation method is used to train Gaussian mixture model (GMM) for the establishment of multi-feature "fingerprint library" of UAV audio signals, and finally the UAV detection and identification is achieved with the feature matching algorithm. The experimental results show that the proposed method can realize the UAV detection and recognition within 150m distance in typical suburban environment with an accuracy rate of 84.4%.