Page 139 - 《应用声学》2021年第3期
P. 139

第 40 卷 第 3 期                                                                       Vol. 40, No. 3
             2021 年 5 月                          Journal of Applied Acoustics                      May, 2021

             ⋄ 研究报告 ⋄



                基于一维卷积神经网络的化爆和地震次声分类                                                                    ∗





                                  谭笑枫 李夕海             †   刘继昊 李广帅 于晓彤



                                                 (火箭军工程大学      西安   710025)

                摘要：次声事件的分类识别方法应用广泛，传统分类方法在很多方面进行了尝试，但由于次声信号具备非线性
                的特点，致使分类难度较大，分类精度不高，这对次声事件的分类工作提出了挑战。针对次声事件中的化学爆
                炸与天然地震信号分类问题，文章构建了一种改进的深度卷积神经网络分类模型用于实现两类次声信号的分
                类。论文采用 “全面禁止核试验条约组织” 官网上收集到的化学爆炸和天然地震次声信号进行数据集的构建，
                使用改进的深度卷积神经网络分别与 BP 网络和一维 LeNet-5 网络进行对比分析。实验结果表明，论文的测试
                识别率能够达到 82.72%，较上述算法有优势。
                关键词：卷积神经网络；次声；分类；识别
                中图法分类号: TN911.7           文献标识码: A          文章编号: 1000-310X(2021)03-0457-11
                DOI: 10.11684/j.issn.1000-310X.2021.03.018



                Classification of chemical explosion and earthquake infrasound based on 1-D
                                            convolutional neural network



                             TAN Xiaofeng LI Xihai LIU Jihao LI Guangshuai YU Xiaotong


                                      (Rocket Force University of Engineering, Xi’an 710025, China)
                 Abstract: The classification and recognition methods of infrasound events have been widely used, and the
                 traditional classification methods have been tried in many aspects, but because of the nonlinear characteristics
                 of infrasound signals, the classification is difficult and the classification accuracy is not high, which poses a
                 challenge to the classification of infrasound events. Aiming at the problem of chemical explosion and natural
                 earthquake signal classification in infrasound events, an improved deep convolutional neural network classifi-
                 cation model is built to realize the classification of two kinds of infrasound signals. In this paper, the chemical
                 explosion and natural earthquake infrasound signals collected on the official website of the Comprehensive
                 Nuclear-Test-Ban Treaty Organization (CTBTO) are used to construct the data set, and the improved deep
                 convolutional neural network is used for comparative analysis with BP network, one-dimensional Lenet-5 net-
                 work. The experimental results show that the test paper recognition rate can reach 82.72%, which is superior
                 to the above algorithm.
                 Keywords: Convolutional neural network; Infrasound; Classification; Recognition





             2020-08-18 收稿; 2021-01-24 定稿
             国家自然科学基金项目 (41774156, 41374154)
             ∗
             作者简介: 谭笑枫 (1993– ), 男, 重庆沙坪坝人, 硕士研究生, 研究方向: 声学信号处理。
             †  通信作者 E-mail: xihai_li@163.com
   134   135   136   137   138   139   140   141   142   143   144