Recognition Method for Underwater Acoustic Target Based on DCGAN and DenseNet
Yingjie Gao, Yuechao Chen, F. Wang, Yalong He
Abstract
The scarcity and access difficulty of labeled underwater acoustic samples have created a bottleneck in introducing deep learning methods into recognition tasks of underwater acoustic targets. In this paper, a recognition method based on the combination of Deep Convolutional Generative Adversarial Network (DCGAN) and Densely Connected Convolutional Networks (DenseNet) for underwater acoustic targets is proposed aiming at these problems. On the basis of meeting the adaption requirements of the deep learning model for the input form, the sample set of wavelet time-frequency graph for the underwater acoustic target was constructed, combined with the prior knowledge of conventional sonar signal processing. The DCGAN model for generation of underwater acoustic sample and the DenseNet model for recognition of underwater acoustic target are designed, and the quality of generated samples is optimized through three stages of iterative training, thus expanding the training set, and improving the recognition effect of underwater acoustic target.