UACE2017 Proceedings: Classification of Whale Calls based on Transfer Learning and Convolutional Neural Network

Back


  • Session:
    Towards Automatic Target Recognition. Detection, Classification and Modelling
  • Paper:
    Classification of Whale Calls based on Transfer Learning and Convolutional Neural Network
  • Author(s):
    Hao Yue, Dezhi Wang, Lilun Zhang, Yanqun Wu, Changchun Bao
  • Abstract:
    As a primary communication method, whale vocal calls contain valuable information and abundant characteristics that are important for recognition and classification. However, subject to large variations of call types, there is still a great challenge to accurately categorize the different whale species or subpopulations. In this study, an effective method of transfer learning based on a data-driven machine-learning approach i.e. Convolutional Neural Network (CNN) is developed to extract the significant features of whale calls and classify the different whale categories from a large open-source acoustic dataset recorded by audio sensors carried by whales .The results show that the proposed method can achieve 97.04% and 91.47% in accuracy respectively to categorize the calls into the two whale species and the four whale subpopulations. The phylogeny graph is also produced to illustrate the similarities between the whale subpopulations. Moreover, all the results are carefully compared with those obtained by using the Wndchrm scheme and the Fisher discriminant scores on the same dataset.
  •   Download the full paper

Contact details

  • Contact person:
    Mr Hao Yue
  • e-mail:
  • Affiliation:
    Academy of Marine Science and Engineering, National University of Defense Technology, Changsha, China
  • Country:
    China

Back