兵器辐射噪声信号的非线性特性及其在被动声信号分类中的应用

The nonlinearity of weapon-radiated noise and its application for the classification of passive acoustic signal

首先利用双谱分析检验了兵器辐射噪声信号的非高斯、非线性特性，结果表明准确的兵器辐射噪声信号模型应为非线性模型。然后，提出了利用一种双线性参数模型对信号进行预处理，并给出了参数估计算法。最后，利用提取的双线性参数模型的系数建立了一积隐含马尔可夫模型，对声信号进行分类。分类结果表明，基于双线性参数模型的积隐含马尔可夫模型分类器，在性能上优于传统的基于AR参数模型的隐含马尔可夫模型分类器，在被动声信号分类中具有良好的应用前景。

The non-gaussianity and nonlinearity of the weapon-radiated noise (WRN) are firstly tested using bispectral analysis, the results show that the model of WRN signals should be nonlinear. Then, a bilinear model (BLM) is presented, which is used to preprocess the WRN signals, and the parameter estimation algorithm is also given. Finally, a product hidden markov model (PHMM ) classifier is builded using the BLM coefficients to classify the acoustic signals. The results show that the BLMPHHM classifier are superior to traditional AR-HMM classifier and have great potentials in the field of passive acoustic signal classification.