基于Diffusion Model的水声数据增强方法

Underwater Acoustic Data Enhancement Method Based on Diffusion Model

摘要:

深度学习需要大量数据对网络模型进行训练才能达到良好的识别效果，而水声数据的数量较少且难以获取。为了解决这一问题，采用了一种基于扩散模型的水声数据增强方法，使用海上试验采集的水声数据进行训练，构建数据生成模型，生成高质量的数据样本，扩充训练数据集。实验结果表明，使用AlexNet作为分类模型进行训练时，该方法相较于传统数据增强方法能够将识别错误率降低约30%，验证了方法的有效性。

Abstract:

Deep learning requires a large amount of data to train network models in order to achieve good recognition performance, while underwater acoustic data is scarce and difficult to obtain. To solve this problem, a diffusion model-based underwater acoustic data augmentation method was adopted, which uses underwater acoustic data collected from offshore experiments for training, constructs a data generation model, generates high-quality data samples, and expands the training dataset. The experimental results show that when using AlexNet as a classification model for training, this method can reduce the recognition error rate by about 30% compared to traditional data augmentation methods, verifying the effectiveness of the method.