Noise Source Recognition Based on Two-Level Architecture Neural Network Ensemble for Incremental Learning

In this paper we propose a two-level architecture ensemble classifier with incremental learning ability for solving the problem of limited experimental sample of the underwater vehicle machinery noise source. The first-level ensemble classifier aims to improve the generalization performance. The second-level ensemble aims to enable the classifier incremental learning. Experimental result shows two-level architecture ensemble classifier has higher accuracy and generalization than traditional classifier, it can overcome the short of wasting time and resources as traditional classifier learn new category that need to reuse all original training data. The two-level architecture ensemble classifier also has incremental learning ability which is important to underwater vehicle machinery noise source recognition actually.