可增量学习的水下航行器噪声源识别中聚类算法研究

水下航行器的噪声源识别具有训练样本有限,存在偶发或突变噪声源等特点。本文针对这些特点,在具有增量学习能力的水下航行器的噪声源识别系统架构下,提出了一种参数自适应可调的基于密度的聚类算法。实验表明,该算法可以有效避免基于密度的聚类算法的参数敏感性对聚类结果的不良影响,在无监督情况下对水下航行器的机械噪声源样本进行有效聚类。通过该聚类算法标注后的样本可直接作为具有增量学习结构的分类器的训练样本,节省了时间和系统开销。

Research on the Clustering Algorithm of the Class-Incremental Learning Model for Underwater Vehicle Noise Source Recognition

The  underwater vehicle machinery noise source recognition features that  the training samples is limited and have abrupt noise samples. Based on these characteristics,this paper proposes a density based algorithm which is parameter adjustable. And this novel algorithm is an  important component of the underwater vehicle machinery noise source recognition system with incremental learning ability. The experimental results show the new algorithm can avoid the parameter sensitivity of DBSCAN. Labeled samples by this algorithm can directly be used as the classifier training samples,saving lots of time and system resources.