基于ICA和HMM的低空声目标识别方法

An approach to low altitude passive acoustic target recognition based on ICA and HMM

提出了一种基于独立分量分析(ICA)和隐马尔可夫模型(HMM)的战场声目标识别方法。针对战场环境下声信号复杂多变,提取目标信号特征难的特点,该方法先利用基于独立分量分析的盲源分离分解混叠信号,再从分离信号中得到更能反应声音特性的Mel倒谱系数作为识别战场低空目标的特征参数;利用隐马尔可夫过程具有很强的表征时变信号的能力来表现声信号随时间变化呈现出的模式演变现象,建立隐马尔可夫模型(HMM)。实际数据的识别分析结果表明了该方法的准确性与有效性。

An approach to identifying low altitude passive acoustic target in battlefield is proposed.Based on Independent Component Analysis（ICA）,the noise is removed from the acoustic signal.Mel-frequency Cepstrum Coefficients（MFCC） are extracted as characteristic parameters in the system.Due to better performance in representing the time-variant signal,the Hidden Markov Models（HMM） are employed to simulate the model change of the sound signals with time.K-means algorithm is used as cluster MFCC to produce training and identifying eigenvector.Simulation results indicate this approach＇s effectiveness in target recognition.