无人机识别的音频特征提取方法

摘要

针对无人机飞行时噪声产生的机理,分别选取基于经验模态分解(EMD)的能量比以及梅尔频率倒谱系数(MFCC)的特征提取算法实现无人机声信号的特征提取,并引用主成分分析(PCA)方法对特征集进行降维融合处理.最后选择矢量量化方法(VQ)作为分类器对不同类型的无人机目标进行分类与识别.实验结果表明特征融合后的分类性能要好于基于单一特征的分类性能,该方法较好地体现不同类型无人机之间的差异,分类结果准确率较高,具有良好的稳定性.%The mechanism of noise generation of drone flight is studied. The feature extraction method based on empirical mode decomposition (EMD) energy ratio and Mel-frequency reciprocal spectrum coefficient (MFRSC) is proposed. The acoustic signal extraction of the drone flight is realized. Then, the principal component analysis (PCA) method is used to combine the feature and reduce the dimension of the feature. Finally, the vector quantification (VQ) is chosen as a classifier to classify and identify the types of drone targets. The experimental results show that the classified performance of the combined feature is better than that of the single feature.The method can reflect the difference between the types of drones and have high accuracy and good stability for classification of features.