Page 26 - 《应用声学》2021年第4期
P. 26

第 40 卷 第 4 期                                                                       Vol. 40, No. 4
             2021 年 7 月                          Journal of Applied Acoustics                      July, 2021

             ⋄ 研究报告 ⋄


                         基于Gammatone滤波器组时频谱和

                               卷积神经网络的海底底质分类                                            ∗




                                            逄 岩     1,2  许 枫     1†  刘 佳     1


                                              (1  中国科学院声学研究所      北京   100190)
                                                (2  中国科学院大学     北京   100049)
                摘要：为了有效利用海底底质信号完成海底底质的分类识别，该文提出一种将深度学习方法和底质信号相结
                合实现底质分类识别的方法。首先利用 Gammatone 滤波器组计算底质侧扫图像信号的时频谱，然后通过卷
                积神经网络对得到的时频谱进行分类识别完成底质分类。利用加利福尼亚州 Scott Creek 近海采集的侧扫声
                呐图像数据进行数据分析，结果表明应用该方法的底质分类准确率平均达到 99.15%，相对于利用分类器分类
                人工提取的底质分类特征，分类性能更加优越；同时利用该方法处理海上试验数据，结果证明该方法具有一定
                的泛化能力。该文研究结果对实际的海底底质分类具有一定参考意义。
                关键词：底质分类；Gammatone 滤波器组；时频分析；时频谱；卷积神经网络
                中图法分类号: O427.9          文献标识码: A          文章编号: 1000-310X(2021)04-0510-08
                DOI: 10.11684/j.issn.1000-310X.2021.04.003


              Seabed sediment classification based on gammatone filter banks time-frequency

                                   spectrum and convolutional neural networks

                                            PANG Yan  1,2  XU Feng 1  LIU Jia 1

                               (1  Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China)
                                   (2  University of Chinese Academy of Sciences, Beijing 100049, China)

                 Abstract: In order to effectively use sea bottom sediment signal to accomplish the classification and recognition
                 of the sediments, a method of combining the deep learning and the sediment signal to achieve the classification
                 and identification of the sea bottom sediment is proposed in this paper. First, the Gammatone filter banks is
                 used to calculate the time-frequency spectrum of sediments side scan sonar image signals. In the end, using a
                 CNN model to classify the time-frequency spectrum calculated by Gammatone filter banks. The results of data
                 analysis with side scan sonar image data collected offshore of Scott Creek, California show that the classification
                 and recognition accuracy of sediments by this method can averagely reach 99.15%, which is superior to using
                 classifiers to classify sediment classification features manually extracted in classification performance, and the
                 results of using this method to process the sea trial data show the means proposed by this paper has a certain
                 generalization ability. The results of this study have specific reference significance for actual seabed sediments
                 classification.
                 Keywords: Sediments classification; Gammatone filter banks; Time-frequency analysis; Time-frequency
                 spectrum; Convolutional neural networks


             2020-08-30 收稿; 2020-12-09 定稿
             国家自然科学基金资助项目 (11404365)
             ∗
             作者简介: 逄岩 (1996– ), 男, 山东青岛人, 博士研究生, 研究方向: 水声信号处理。
              通信作者 E-mail: xf@mail.ioa.ac.cn
             †
   21   22   23   24   25   26   27   28   29   30   31