Page 189 - 《应用声学》2023年第3期
P. 189
第 42 卷 第 3 期 Vol. 42, No. 3
2023 年 5 月 Journal of Applied Acoustics May, 2023
⋄ 研究报告 ⋄
基于非负矩阵分解的次声信号分类方法 ∗
孟子轩 1,2,3 程 巍 1,2,3 张天予 1,2,3 吕 君 1,2 滕鹏晓 1,2†
(1 中国科学院声学研究所 北京 100190)
(2 中国科学院噪声与振动重点实验室 北京 100190)
(3 中国科学院大学 北京 100049)
摘要:在中国科学院声学研究所大气次声波观察网实地采集的爆炸、地震、闪电、再入 4 类次声事件 105 组阵
列数据集的基础上,提出应用非负矩阵分解的特征提取方法,对次声信号的自动分类方法进行了研究。针对特
征设计过程复杂的问题,该方法使用非负矩阵分解自动挖掘目标信号的隐含结构作为特征。将此特征作为支
持向量机和卷积神经网络输入进行分类,以提高特征设计的效率与分类的识别准确率。研究结果指出,在测试
集上的平均识别准确率达到了 83.13%,相对于传统方法,简化了特征设计过程,并取得更好的分类结果。
关键词:次声信号;特征提取;非负矩阵分解;信号分类
中图法分类号: O425+.3 文献标识码: A 文章编号: 1000-310X(2023)03-0627-10
DOI: 10.11684/j.issn.1000-310X.2023.03.022
Study on classification of infrasound signals based on nonnegative matrix
factorization
MENG Zixuan 1,2,3 CHENG Wei 1,2,3 ZHANG Tianyu 1,2,3 LYU Jun 1,2 TENG Pengxiao 1,2
(1 Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China)
(2 Key Laboratory of Noise and Vibration Research, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China)
(3 University of Chinese Academy of Sciences, Beijing 100049, China)
Abstract: On the basis of 105 sets of array data from explosion, earthquake, lightning and rocket reentry
infrasonic events, which collected by the infrasonic observation website of the Institute of Acoustics, Chinese
Academy of Sciences, the automatic classification method of non-negative matrix decomposition is studied. For
the complex problem of the feature design process, this method uses non-negative matrix decomposition to
automatically mine the implicit structure of the target signal as features. Using this feature for the classification
experiment as the input of support vector machine(SVM) and convolutional neural network improves the
efficiency of feature design and the identification accuracy of classification. The results point out that the
average recognition accuracy on the test set reaches 83.13%, which simplifies the feature design process and
achieves better classification results relative to the traditional methods.
Keywords: Infrasound signal; Feature extraction; Non-negative matrix factorization; Signal classification
2022-02-28 收稿; 2022-03-29 定稿
国家自然科学基金项目 (11774372, 11874389)
∗
作者简介: 孟子轩 (1996– ), 男, 山东潍坊人, 硕士研究生, 研究方向: 次声信号处理。
† 通信作者 E-mail: px.teng@mail.ioa.ac.cn