Scholarlink

<p>Ship radiated noise (SRN) is an important source of information for passive sonar systems to identify ship targets. Passive sonar detection and recognition of underwater targets have become increasingly difficult due to the continuous improvement of the ability of underwater acoustic targets to reduce shock and noise. To address this issue, this paper proposes a method for underwater acoustic target recognition that combines the time-domain, frequency-domain, and entropy features of radiated noise. The entropy features exhibit low computational complexity and strong noise robustness, making them highly suitable for quantifying the complexity of SRN signals. Unlike existing studies that focus only on the fusion of entropy metrics, we proposed a multi-frequency-band entropy-based feature combination, which significantly enhances noise robustness while reducing computational complexity. The proposed method combines the permutation entropy (PE) from full frequency band, envelope entropy (EE) from 4–8 kHz frequency band, and spectral entropy (SE) from 0.01–0.1 kHz frequency band with time-domain features (mean and variance) and frequency-domain features (spectral centroid, kurtosis, and variance). The above features were extracted from the DeepShip dataset and input into different classifiers, including random forest (RF), AdaBoost, convolutional neural networks (CNN) and other machine learning classifiers, to verify the effectiveness of the features. The experimental results showed that the proposed method achieved the recognition accuracy of 85.92% and 86.59% on the RF and AdaBoost models, respectively. Although the CNN model was included only as a structural baseline, it still outperformed mainstream deep learning models with the accuracy of 80.71%. In addition, ocean background noise interference was introduced into the experimental data to verify the robustness of the proposed method and compared the results of the proposed method with existing mainstream methods. The results showed that the proposed method exhibited a better recognition performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Entropy-based feature combination method for ship radiated noise in underwater acoustics target recognition

  • 期刊论文

  • 发表日期：

    2026年2月1日

查看全文
  • Chenqi Jiang,
  • Wei Zheng,
  • Biao Wang,
  • Zhenkai Zhang

摘要

Ship radiated noise (SRN) is an important source of information for passive sonar systems to identify ship targets. Passive sonar detection and recognition of underwater targets have become increasingly difficult due to the continuous improvement of the ability of underwater acoustic targets to reduce shock and noise. To address this issue, this paper proposes a method for underwater acoustic target recognition that combines the time-domain, frequency-domain, and entropy features of radiated noise. The entropy features exhibit low computational complexity and strong noise robustness, making them highly suitable for quantifying the complexity of SRN signals. Unlike existing studies that focus only on the fusion of entropy metrics, we proposed a multi-frequency-band entropy-based feature combination, which significantly enhances noise robustness while reducing computational complexity. The proposed method combines the permutation entropy (PE) from full frequency band, envelope entropy (EE) from 4–8 kHz frequency band, and spectral entropy (SE) from 0.01–0.1 kHz frequency band with time-domain features (mean and variance) and frequency-domain features (spectral centroid, kurtosis, and variance). The above features were extracted from the DeepShip dataset and input into different classifiers, including random forest (RF), AdaBoost, convolutional neural networks (CNN) and other machine learning classifiers, to verify the effectiveness of the features. The experimental results showed that the proposed method achieved the recognition accuracy of 85.92% and 86.59% on the RF and AdaBoost models, respectively. Although the CNN model was included only as a structural baseline, it still outperformed mainstream deep learning models with the accuracy of 80.71%. In addition, ocean background noise interference was introduced into the experimental data to verify the robustness of the proposed method and compared the results of the proposed method with existing mainstream methods. The results showed that the proposed method exhibited a better recognition performance.