倪俊帅,赵梅,胡长青.基于DNN和改进K-means的船舶辐射噪声开集识别方法[J].声学技术,2022,41(3):382~387 |
基于DNN和改进K-means的船舶辐射噪声开集识别方法 |
DNN and improved K-means based ship noise open set recognition |
投稿时间:2021-06-16 修订日期:2021-07-22 |
DOI:10.16300/j.cnki.1000-3630.2022.03.011 |
中文关键词: 船舶辐射噪声 深度神经网络(DNN) 改进 K-means 开集识别 |
英文关键词: ship radiation noise Deep Neural Network improved K-means open set recognition |
基金项目:水声对抗技术重点实验室开放基金(JCKY2020207CH02)、中国科学院声学研究所自主部署项目资助项目。 |
|
摘要点击次数: 436 |
全文下载次数: 318 |
中文摘要: |
为提高船舶噪声识别系统的性能,实现开集识别,提出了基于深度神经网络(Deep Neural Network,DNN)和改进K-means的船舶辐射噪声开集识别方法。首先,采用Welch功率谱估计方法提取船舶辐射噪声的特征;然后,设计并应用DNN模型进一步提取特征向量;最后,使用改进的K-means模型实现开集识别。在实测数据上进行了实验,结果表明,所提方法能实现船舶辐射噪声开集识别,对于实测数据的平均识别正确率为93.5%,较DNN+K-means++方法提高了6.2个百分点。对实测数据添加实验船发动机噪声或渔船噪声进行实验,结果表明,识别方法在其他船只噪声干扰下具有较好的鲁棒性。 |
英文摘要: |
In order to improve the performance of ship noise recognition system and realize open set recognition, an open set recognition method of ship radiation noise based on Deep Neural Network (DNN) and improved K-means is proposed. First, the Welch power spectrum estimation method is used to extract the characteristics of ship radiation noise. Then, the deep neural network model is designed for further extraction of feature vectors. Finally, the improved K-means model is used to realize open set recognition. Experiments are carried out on the measured data, and the results show that the proposed method can realize the open set recognition of ship radiation noise. The average recognition accuracy for the measured data is 93.5%, which is 6.2% higher than that of the DNN+K-means++ method. After adding experimental ship engine noise or fishing boat noise to the measured data, the experimental results show that the recognition method has good robustness under the interference of other ship noises. |
查看全文
查看/发表评论 下载PDF阅读器 |
关闭 |