基于一维卷积和LSTM网络的端到端水声目标识别
End-to-end acoustic target recognition based on 1D convolutional and LSTM networks
杨康1
作者信息
- 1. 镇江市高等专科学校,江苏 镇江 212003;江苏科技大学,江苏 镇江 212003
- 折叠
摘要
水声目标识别在国防和海洋环境监测等领域具有重要应用.然而,传统的时频域特征提取方法由于信息损失和环境适应性不足,限制了识别性能的提升.为克服这些局限性,文章提出了一种基于一维卷积神经网络(One-dimensional Convolutional Neural Network,1D CNN)与长短时记忆网络(Long Short-term Memory Network,LSTM)相结合的端到端水声目标识别模型(One-dimensional Long Short-term Memory,1DLSTM).该模型直接以原始时域信号为输入,利用1D CNN提取局部特征,通过LSTM捕捉长程依赖关系,有效保留了信号的全局信息.在ShipsEar数据集上的实验结果表明,该模型的识别准确率高达93.91%,为水声目标端到端识别领域提供了一种新思路.
Abstract
Acoustic target recognition plays a crucial role in defense and marine environment monitoring.However,traditional time-frequency domain feature extraction methods often suffer from information loss and inadequate adaptability to varying environments,limiting their recognition performance.To address these limitations,this paper presents an end-to-end acoustic target recognition model(1DLSTM)that combines a one-dimensional convolutional neural network(1D CNN)with a long short-term memory network(LSTM).This model directly processes raw time-domain signals,using the 1D CNN to extract local features and the LSTM to capture long-term dependencies,thereby effectively preserving the global information of the signal.Experimental results on the ShipsEar dataset demonstrate that this model achieves a recognition accuracy of 93.91%,offering a novel approach to end-to-end acoustic target recognition.
关键词
深度学习/水声目标识别/端到端Key words
deep learning/acoustic target recognition/end-to-end引用本文复制引用
出版年
2024
NETL
NSTL
万方数据