子类独立分量分析在声目标识别中的应用

Application of subclass independent component analysis in acoustic target identification

针对已有的特征提取方法在多目标识别中的不足,提出了基于高阶统计分析的独立分量分析法特征提取方法,通过对多种目标的声音信号进行子类特征提取,并应用决策导向无环图支持向量机实现对多目标的有效分类。结果表明该算法在通过声音信号对多目标识别上,具有很好的应用前景。

We propose a new method of multi-targets identification based on independent component analysis （ICA） and decision directed acyclic graph support vector machine （DDAGSVMs）, which overcomes defects of proposed targets identification. Features of the acoustic wave is extracted by ICA based on the higher-order statistics. Support vector machine （SVM） is applied in targets recognition. Finally we describe experiments based on ICA and SVM. The results show that the new method is an intelligent measure to identify acoustic targets, and has broad application prospects.