Hydroacoustic Target Recognition Based on Multidiscriminative Generative Adversarial Network

To address the issue of sample scarcity in hydroacoustic target recognition, a multi-discriminative generative adversarial network (MD-GAN) method is proposed. Firstly, the fixed mapping interval method is adopted to preprocess the ship signal and extract the LOFAR spectrogram of the target. Then the designed MD-GAN is used to perform data enhancement on these spectrograms. By integrating different numbers and configurations of discriminator networks, we generate sample data and evaluate their quality and diversity. Furthermore, we simplify the hidden layers of the VGG16 network to obtain a version with fewer parameters but still excellent classification performance. The effectiveness of the method under small samples is verified by analyzing the impact of different generated samples on the target recognition accuracy. Experimental results show that the recognition accuracy reaches 95.87%, indicating that the method has great potential in the field of hydroacoustic target recognition, especially in the case of limited samples.