为解决目前声目标识别面临的鲁棒性不足问题,提出将流形学习应用到声目标的特征提取中,在经典流形学习算法的基础上,研究讨论了目标声信号频域中存在的低维流形,通过两种实际的地面和低空飞行声目标数据集进行对比识别实验,分析了基于流形学习的声目标特征提取方法的性能,结果表明基于流形学习的特征提取方法可以发现声信号的本质特征,提高了声目标识别系统的准确性和鲁棒性. In order to solve the problem of lack of robustness to the current target recognition, this paper proposes to apply manifold learning to the feature extraction of acoustic targets. On the basis of the classical manifold learning algorithm, the existence of low Dimensional manifolds, two kinds of real ground-based and low-altitude flight target data sets are compared for recognition experiments, and the performance of the method based on manifold learning is analyzed. The results show that the feature extraction method based on manifold learning can find sound The nature of the signal improves the accuracy and robustness of the acoustic target recognition system.