Acoustic Signal Target Recognition Using Improved Clustering Autoencoder

Abstract

According to the small-sample data discretization problem, this paper proposes an acoustic signal target recognition model using improved clustering autoencoder(ICAE) to complete acoustic signal target recognition. Specifically, the clustering loss function of the proposed ICAE is developed to encode and cluster the identity authentication(I-vector), which can solve the large gap between a small amount of target-related data and the poor recognition effect. Moreover, the linear discrimination analysis(LDA) is adopted to project the dataset on the feature subspace that can differentiate the different targets with dimensionality reduction. The experimental results show that the recognition model using the proposed ICAE can achieve better recognition performance and strong adaptability. Compared with other methods, the proposed ICAE in this paper has an obvious clustering effect on a small amount of data.