基于ICA和HMM的战场混叠声目标识别

Identification of Field Mixing Acoustic Targets Based on ICA and HMM

为识别战场混叠声目标,提出一种基于独立分量分析(ICA)的声目标盲分离和隐马尔可夫(HMM)识别的混合声识别方法.建立已知声目标的HMM,实现混叠声目标盲分离,提取的线性预测系数作为声目标识别参数,通过K均值聚类得到训练和识别特征向量,通过Viterbi解码判断声目标的类别.仿真结果表明,ICA分析能有效地分离混叠声目标信号,基于线性预测系数的HMM识别率较高,混合模型识别系统在混叠声目标识别中具有可行性.

To identify mixing acoustic targets, the independent component analysis （ICA） blind separation from mixing acoustic targets was combined with HMM identification. An identification method based on HMM was established to separate mixing acoustic targets through extracting LPC characteristics. The HMM was employed to compute the Viterbi output score, and K-means algorithm was used as cluster LPC coefficients. Targets were identified by ICA-HMM model. Simulation indicates that ICA can separate mixing targets efficiently. HMM identification performance based on LPC hybrid model is good, and it is effective on mixing acoustic targets identification.