基于生成对抗网络的舰船辐射噪声分类方法研究

基于机器学习的舰船目标识别近年来已成为水声信号处理领域的一个重要研究方向，但水声目标信号的获取困难，样本量不足和不均衡的问题很容易导致目标分类模型的识别效果不佳。该文提出一种基于条件卷积生成对抗网络的船舶噪声数据分类方法，该方法利用生成对抗学习理论，生成相比于传统数据增强算法非线性特征更强，特征差异更丰富的伪DEMON调制谱数据来缓解训练样本量不足的问题。之后将传统生成对抗网络中的全连层输出替换成更善于解决小样本问题集成分类器，从而降低分类器对于数据量的依赖程度，进一步提高分类模型性能。最终由基于真实样本的实验结果表明，相比于传统数据增强算法和卷积生成对抗网络，该文方法能够更有效提高在样本不足条件下的模型的分类性能。

Research on Classification Algorithm of Ship Radiated Noise Data Based on Generative Adversarial Network

In recent years, ship target recognition based on machine learning has become an important research direction in the field of underwater acoustic signal processing, but the acquisition of underwater acoustic target signal is difficult, and the problem of insufficient sample size and imbalance leads easily to the poor recognition effect of target classification model. A ship noise data classification method based on Generative Admission-Network (GAN) is proposed in this paper. This method uses generative admission-learning theory to generate pseudo-DEMON modulation spectrum data with stronger nonlinear characteristics and richer feature differences compared with traditional data enhancement algorithms to alleviate the problem of insufficient training sample size. Then, the output of the whole connection layer in the traditional generative adversarial network is replaced by an ensemble classifier which is better at solving the problem of small samples, so as to reduce the dependence of the classifier on the amount of data and improve further the performance of the classification model. Finally, experimental results based on real samples show that, compared with traditional data enhancement algorithms and generative adversarial networks, the proposed method can improve the classification performance of models with insufficient samples more effectively.