Page 17 - 《应用声学》2024年第6期
P. 17

第 43 卷 第 6 期                                                                       Vol. 43, No. 6
             2024 年 11 月                         Journal of Applied Acoustics                 November, 2024

             ⋄ 研究报告 ⋄


                应用原型网络的小样本次声信号分类识别方法                                                                    ∗



               赵子杰      1,2,3  程 巍    1,2   姬培锋     1,2  滕鹏晓      1,2   吕 君     1,2   杨 军    1,2,3†



                                              (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：地震、闪电、火箭发射、爆炸等活动都会伴随着次声信号的产生。为提升次声事件的监测能力，需要对小
                样本的次声信号进行正确分类识别。针对小样本集的次声事件的有效识别问题，结合长短期记忆模型提出了
                一种应用原型网络的次声信号分类方法。使用该方法分别对公开的次声信号数据集和实地采集的地震、爆炸、
                闪电、火箭再入产生的 4 类次声信号进行分类实验。实验结果表明，该方法相对于传统方法，简化了特征提取
                的过程，有效解决了小样本集次声信号的特征分析问题，取得较好的分类结果和泛化效果。
                关键词：次声；小样本；原型网络；长短期记忆模型
                中图法分类号: TN911.7           文献标识码: A          文章编号: 1000-310X(2024)06-1193-10
                DOI: 10.11684/j.issn.1000-310X.2024.06.002


                      A method for classification of few-shot infrasound signals applying

                                                  prototype network

                            ZHAO Zijie 1,2,3  CHENG Wei  1,2  JI Peifeng 1,2  TENG Pengxiao 1,2

                                               LYU Jun 1,2   YANG Jun  1,2,3

                               (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: Events such as earthquakes, lightning, rocket launches, and explosions are accompanied by infra-
                 sound signals. In order to improve the monitoring capability of infrasound events, it is necessary to correctly
                 classify small samples of infrasound signals. For the problem of effective identification of infrasound events
                 with small samples and variable duration, a classification method of infrasound signals applying prototype net-
                 work is proposed in combination with a long and short-term memory model. The method is used to conduct
                 classification experiments on publicly available infrasound signal datasets and four types of infrasound signals
                 generated by earthquakes, explosions, lightning, and rocket re-entry collected in the field. The experimental
                 results show that the method simplifies the process of feature extraction and effectively solves the problem of
                 feature analysis of variable duration infrasound signals compared with the traditional method, and achieves
                 better classification results and generalization effects.
                 Keywords: Infrasound; Few-shot; Prototypical network; Long short-term memory


             2023-02-28 收稿; 2023-04-07 定稿
             国家自然科学基金项目 (11874389)
             ∗
             作者简介: 赵子杰 (1997– ), 男, 河北衡水人, 硕士研究生, 研究方向: 信号与信息处理。
             † 通信作者 E-mail: jyang@mail.ioa.ac.cn
   12   13   14   15   16   17   18   19   20   21   22